Foundations and Trends

R
in Optimization
Vol. 1, No. 2 (2013) 70–122
c 2013 M. Kraning, E. Chu, J. Lavaei, and S. Boyd

DOI: xxx

Dynamic Network Energy Management via

Proximal Message Passing

Matt Kraning Eric Chu

Stanford University Stanford University
[email protected] [email protected]
Javad Lavaei Stephen Boyd
Columbia University Stanford University
[email protected] [email protected]
 Contents

1 Introduction 70
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
1.2 Related work . . . . . . . . . . . . . . . . . . . . . . . . . 73
1.3 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

2 Network Model 76
2.1 Formal definition and notation . . . . . . . . . . . . . . . 76
2.2 Dynamic optimal power flow problem . . . . . . . . . . . . 78
2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 79
2.4 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

3 Device Examples 83
3.1 Generators . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.2 Transmission lines . . . . . . . . . . . . . . . . . . . . . . 84
3.3 Converters and interface devices . . . . . . . . . . . . . . 86
3.4 Storage devices . . . . . . . . . . . . . . . . . . . . . . . 87
3.5 Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

4 Convexity 91
4.1 Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2 Relaxations . . . . . . . . . . . . . . . . . . . . . . . . . . 92

ii
 iii

5 Proximal Message Passing 95

5.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.2 Convergence . . . . . . . . . . . . . . . . . . . . . . . . . 98
5.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . 101

6 Numerical Examples 103

6.1 Network topology . . . . . . . . . . . . . . . . . . . . . . 103
6.2 Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.3 Serial multithreaded implementation . . . . . . . . . . . . 107
6.4 Peer-to-peer implementation . . . . . . . . . . . . . . . . 107
6.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

7 Extensions 111
7.1 Closed-loop control . . . . . . . . . . . . . . . . . . . . . 111
7.2 Security constrained optimal power flow . . . . . . . . . . 113
7.3 Hierarchical models and virtualized devices . . . . . . . . . 113
7.4 Local stopping criteria and ρ updates . . . . . . . . . . . . 115

8 Conclusion 116
Abstract
We consider a network of devices, such as generators, fixed loads, de-
ferrable loads, and storage devices, each with its own dynamic con-
straints and objective, connected by AC and DC lines. The problem is
to minimize the total network objective subject to the device and line
constraints over a time horizon. This is a large optimization problem
with variables for consumption or generation for each device, power
flow for each line, and voltage phase angles at AC buses in each period.
We develop a decentralized method for solving this problem called
proximal message passing. The method is iterative: At each step, each
device exchanges simple messages with its neighbors in the network
and then solves its own optimization problem, minimizing its own ob-
jective function, augmented by a term determined by the messages it
has received. We show that this message passing method converges to
a solution when the device objective and constraints are convex. The
method is completely decentralized, and needs no global coordination
other than synchronizing iterations; the problems to be solved by each
device can typically be solved extremely efficiently and in parallel.
The proximal message passing method is fast enough that even a se-
rial implementation can solve substantial problems in reasonable time
frames. We report results for several numerical experiments, demon-
strating the method’s speed and scaling, including the solution of a
problem instance with over 30 million variables in 5 minutes for a serial
implementation; with decentralized computing, the solve time would be
less than one second.
 1
Introduction

1.1 Overview
A traditional power grid is operated by solving a number of opti-
mization problems. At the transmission level, these problems include
unit commitment, economic dispatch, optimal power flow (OPF), and
security-constrained OPF (SC-OPF). At the distribution level, these
problems include loss minimization and reactive power compensation.
With the exception of the SC-OPF, these optimization problems are
static with a modest number of variables (often less than 10000), and
are solved on time scales of 5 minutes or more. However, the operation
of next generation electric grids (i.e., smart grids) will rely critically on
solving large-scale, dynamic optimization problems involving hundreds
of thousands of devices jointly optimizing tens to hundreds of millions
of variables, on the order of seconds rather than minutes [16, 41]. More
precisely, the distribution level of a smart grid will include various types
of active dynamic devices, such as distributed generators based on solar
and wind, batteries, deferrable loads, curtailable loads, and electric ve-
hicles, whose control and scheduling amount to a very complex power
management problem [59, 9].
In this paper, we consider a general problem, which we call the

70
1.1. Overview 71

dynamic optimal power flow problem (D-OPF), in which dynamic de-

vices are connected by both AC and DC lines, and the goal is to jointly
minimize a network objective subject to local constraints on the devices
and lines. The network objective is the sum of the objective functions of
the devices. These objective functions extend over a given time horizon
and encode operating costs such as fuel consumption and constraints
such as limits on power generation or consumption. In addition, the
objective functions encode dynamic objectives and constraints such as
limits on ramp rates for generators or charging and capacity limits for
storage devices. The variables for each device consist of its consumption
or generation in each time period and can also include local variables
which represent internal states of the device over time, such as the state
of charge of a storage device.
When all device objective functions and line constraints are convex,
D-OPF is a convex optimization problem, which can in principle be
solved efficiently [7]. If not all device objective functions are convex, we
can solve a relaxed form of the D-OPF which can be used to find good,
local solutions to the D-OPF. The optimal value of the relaxed D-OPF
also gives a lower bound for the optimal value of the D-OPF which can
be used to evaluate the suboptimality of a local solution, or, when the
local solution has the same value, as a certificate of global optimality.
For any network, the corresponding D-OPF contains at least as
many variables as the number of devices and lines multiplied by the
length of the time horizon. For large networks with hundreds of thou-
sands of devices and a time horizon with tens or hundreds of time peri-
ods, the extremely large number of variables present in the correspond-
ing D-OPF makes solving it in a centralized fashion computationally
impractical, even when all device objective functions are convex.
We propose a decentralized optimization method which efficiently
solves the D-OPF by distributing computation across every device in
the network. This method, which we call proximal message passing,
is iterative: At each iteration, every device passes simple messages to
its neighbors and then solves an optimization problem that minimizes
the sum of its own objective function and a simple regularization term
that only depends on the messages it received from its neighbors in the
72 Introduction

previous iteration. As a result, the only non-local coordination needed

between devices for proximal message passing is synchronizing itera-
tions. When all device objective functions are convex, we show that
proximal message passing converges to a solution of the D-OPF.
Our algorithm can be used several ways. It can be implemented in
a traditional way on a single computer or cluster by collecting all the
device constraints and objectives. We will demonstrate this use with
an implementation that runs on a single 32-core computer with hy-
perthreading (64 independent threads). A more interesting use is in a
peer-to-peer architecture, in which each device contains its own pro-
cessor, which carries out the required local dynamic optimization and
exchanges messages with its neighbors on the network. In this setting,
the devices do not need to divulge their objectives or constraints; they
only need to support a simple protocol for interacting with their neigh-
bors. Our algorithm ensures that the network power flows and AC bus
phase angles will converge to their optimal values, even though each
device has very little information about the rest of the network, and
only exchanges limited messages with its immediate neighbors.
Due to recent advances in convex optimization [61, 46, 47], in many
cases the optimization problems that each device solves in each it-
eration of proximal message passing can be executed at millisecond
or even microsecond time-scales on inexpensive, embedded processors.
Since this execution can happen in parallel across all devices, the entire
network can execute proximal message passing at kilohertz rates. We
present a series of numerical examples to illustrate this fact by using
proximal message passing to solve instances of the D-OPF with over
30 million variables serially in 5 minutes. Using decentralized comput-
ing, the solve time would be essentially independent of the size of the
network and require just a fraction of a second.
We note that although a primary application for proximal message
passing is power management, it can easily be adapted to more general
resource allocation and graph-structured optimization problems [51, 2].
1.2. Related work 73

1.2 Related work

The use of optimization in power systems dates back to the 1920s and
has traditionally concerned the optimal dispatch problem [22], which
aims to find the lowest cost method for generating and delivering power
to consumers, subject to physical generator constraints. With the ad-
vent of computer and communication networks, many different ways
to numerically solve this problem have been proposed [62] and more
sophisticated variants of optimal dispatch have been introduced, such
as OPF, economic dispatch, and dynamic dispatch [12], which extend
optimal dispatch to include various reliability and dynamic constraints.
For reviews of optimal and economic dispatch as well as general power
systems, see [4] and the book and review papers cited above.
When modeling AC power flow, the D-OPF is a dynamic version of
the OPF [8], extending the latter to include many more types of devices
such as storage units. Recent smart grid research has focused on the
ability of storage devices to cut costs and catalyze the consumption of
variable and intermittent renewables in the future energy market [23,
44, 13, 48]. With D-OPF, these storage concerns are directly addressed
and modeled in the problem formulation with the introduction of a time
horizon and coupling constraints between variables across periods.
Distributed optimization methods are naturally applied to power
networks given the graph-structured nature of the transmission and
distribution networks. There is an extensive literature on distributed
optimization methods, dating back to the early 1960s. The prototypi-
cal example is dual decomposition [14, 17], which is based on solving
the dual problem by a gradient method. In each iteration, all devices
optimize their local (primal) variables based on current prices (dual
variables). Then the dual variables are updated to account for imbal-
ances in supply and demand, with the goal being to determine prices
for which supply equals demand.
Examples of distributed algorithms in the power systems litera-
ture include two phase procedures that resemble a single iteration of
dual decomposition. In the first phase, dynamic prices are set over a
given time horizon (usually hourly over the following 24 hours) by some
mechanism (e.g., centrally by an ISO [28, 29], or through information
74 Introduction

aggregation in a market [57]). In the second phase, these prices allow

individual devices to jointly optimize their power flows with minimal
(if any) additional coordination over the time horizon. More recently,
building on the work of [39], a distributed algorithm was proposed [38]
to solve the dual OPF using a standard dual decomposition on subsys-
tems that are maximal cliques of the power network.
Dual decomposition methods are not robust, requiring many tech-
nical conditions, such as strict convexity and finiteness of all local cost
functions, for both theoretical and practical convergence to optimality.
One way to loosen the technical conditions is to use an augmented La-
grangian [25, 49, 5], resulting in the method of multipliers. This subtle
change allows the method of multipliers to converge under mild tech-
nical conditions, even when the local (convex) cost functions are not
strictly convex or necessarily finite. However, this method has the dis-
advantage of no longer being separable across subsystems. To achieve
both separability and robustness for distributed optimization, we can
instead use the alternating direction method of multipliers (ADMM)
[21, 20, 15, 6]. ADMM is very closely related to many other algorithms,
and is identical to Douglas-Rachford operator splitting; see, e.g., the
discussion in [6, §3.5].
Augmented Lagrangian methods (including ADMM) have previ-
ously been applied to the study of power systems with static, single
period objective functions on a small number of distributed subsystems,
each representing regional power generation and consumption [35]. For
an overview of related decomposition methods applied to power flow
problems, we direct the reader to [36, 1] and the references therein.
The proximal message passing decentralized power scheduling
method is similar in spirit to flow control on a communication network,
where each source modulates its sending rate based only on information
about the number of un-acknowledged packets; if the network state re-
mains constant, the flows converge to levels that satisfy the constraints
and maximize a total utility function [33, 42]. In Internet flow control,
this is called end-point control, since flows are controlled (mostly) by
devices on the edges of the network. A decentralized proximal message
passing method is closer to local control, since decision making is based
1.3. Outline 75

only on interaction with neighbors on the network. Another difference

is that the messages our method passes between devices are virtual, and
not actual energy flows. (Once converged, of course, they can become
actual energy flows.)

1.3 Outline
The rest of this paper is organized as follows. In chapter 2 we give the
formal definition of our network model. In chapter 3 we give examples
of how to model specific devices such as generators, deferrable loads
and energy storage systems in our formal framework. In Chapter 4, we
describe the role that convexity plays in the D-OPF and introduce the
idea of convex relaxations as a tool to find solutions to the D-OPF in
the presence of non-convex device objective functions. In Chapter 5 we
derive the proximal message passing equations. In Chapter 6 we present
a series of numerical examples, and in Chapter 7 we discuss how our
framework can be extended to include use cases we do not explicitly
cover in this paper.
 2
Network Model

We begin with an abstract definition of our network model and the

dynamic optimal power flow problem, and the compact notation we
use to describe it. We then give some discussion and an example to
illustrate how the model is used to describe a real power network.

2.1 Formal definition and notation

A network consists of a finite set of terminals T , a finite set of devices
D, and a finite set of nets N . The sets D and N are both partitions
of T . Thus, each device and each net has a set of terminals associated
with it, and each terminal is associated with exactly one device and
exactly one net. Equivalently, a network can be defined as a bipartite
graph with one set of vertices given by devices, the other set of vertices
given by nets, and edges given by terminals — very similar in nature
to ‘normal realizations’ [19] of graphs in coding theory.
Each terminal t ∈ T has a type, either AC or DC, corresponding to
the type of power that flows through the terminal. The set of terminals
can be partitioned by type into the sets T dc and T ac , which represent
the set of all terminals of type DC and AC, respectively. A terminal of
either type has an associated power schedule pt = (pt (1), . . . , pt (T )) ∈

76
2.1. Formal definition and notation 77

RT , where T is a given time horizon. Here, pt (τ ) is the amount of power

consumed by device d in time period τ through terminal t, where t is
associated with d. When pt (τ ) < 0, −pt (τ ) is the energy generated
by device d through terminal t in time period τ . (For AC terminals,
pt is the real power flow; we do not consider reactive power in this
paper.) In addition to (real) power schedules, AC terminals t ∈ T ac also
have phase schedules θt = (θt (1), . . . , θt (T )) ∈ RT , which represent the
absolute voltage phase angles for terminal t over time. (DC terminals
are not associated with voltage phase angles.)
We use a simple method for indexing quantities such as power sched-
ules that are associated with each terminal and vary over time. For
devices d ∈ D, we use ‘d’ to refer to both the device itself as well as
the set of terminals associated with it, i.e., we say t ∈ d if terminal t is
associated with device d. The set of all power schedules associated with
device d is denoted by pd = {pt | t ∈ d}, which we can associate with a
|d| × T matrix. We use the same notation for nets as we do for devices.
The set of all terminal power schedules is denoted by p = {pt | t ∈ T },
which we can associate with a |T | × T matrix. For other quantities that
are associated with each terminal (such as phase schedules), we use an
identical notation to power schedules, i.e., θd = {θt | t ∈ d} is the set of
phase schedules associate with device d (with an identical notation for
nets), and the set of all phase schedules is denoted by θ = {θt | t ∈ T }.
Each device d contains a set of |d| terminals and has an associated
ac
objective function fd : R|d|×T × R|d |×T → R ∪ {+∞}, where dac = {t |
t ∈ d ∩ T ac } is the set of all AC terminals associated with device d, and
we set fd (pd , θd ) = ∞ to encode constraints on the power and phase
schedules for the device. When fd (pd , θd ) < ∞, we say that (pd , θd )
are a set of realizable power and phase schedules for device d, and we
interpret fd (pd , θd ) as the cost (or revenue, if negative) to device d for
operating according to power schedule pd and phase schedule θd .
Similarly, each net n ∈ N contains a set of |n| terminals, all of which
are required to have the same type. (We will model AC–DC conversion
using devices.) We refer to nets containing AC terminals as AC nets
and nets containing DC terminals as DC nets. Nets are lossless energy
carriers which constrain the power schedules (and phase schedules in
78 Network Model

the case of AC nets) of their constituent terminals: we require power

balance in each time period, which is represented by the constraints
X
pt (τ ) = 0, τ = 1, . . . , T, (2.1)
t∈n
for each n ∈ N . In addition to power balance, each AC net imposes
the phase consistency constraints
θt1 (τ ) = · · · = θt|n| (τ ), τ = 1, . . . , T, (2.2)
where n = {t1 , . . . , t|n| }. In other words, in each time period the power
flows on each net balance, and all terminals on the same AC net have
the same phase.
We define the average net power imbalance p̄ : T → RT , as
1 X
p̄t = pt′ , (2.3)
|n| t′ ∈n
where t ∈ n, i.e., terminal t is associated with net n. In other words,
p̄t (τ ) is the average power schedule of all terminals associated with the
same net as terminal t at time τ . We overload this notation for devices
by defining p̄d = {p̄t | t ∈ d}. Using an identical notation for nets,
we can see that p̄n simply contains |n| copies of the average net power
imbalance for net n. The net power balance constraint for all terminals
can be expressed as p̄ = 0.
For AC terminals, we define the phase residual θ̃ : T ac → RT as
1 X
θ̃t = θt − θt′ = θt − θ̄t ,
|n| t′ ∈n

where t ∈ n and n is an AC net. In other words, θ̃t (τ ) is the difference

between the phase angle of terminal t and the average phase angle of
all terminals attached to net n, at time τ . As with the average power
imbalance, we overload this notation for devices by defining θ̃d = {θ̃t |
t ∈ d ∩ T ac } with a similar notation for nets. The phase consistency
constraint for all AC terminals can be expressed as θ̃ = 0.

2.2 Dynamic optimal power flow problem

We say that a set of power and phase schedules p : T → RT , θ :
T ac → RT is feasible if fd (pd , θd ) < ∞ for all d ∈ D (i.e., all devices’
2.3. Discussion 79

power and phase schedules are realizable), and both p̄ = 0 and θ̃ = 0

(i.e., power balance and phase consistency holds across all nets). We
define the network objective as f (p, θ) = d∈D fd (pd , θd ). The dynamic
P

optimal power flow problem (D-OPF) is

minimize f (p, θ)
(2.4)
subject to p̄ = 0, θ̃ = 0,

with variables p : T → RT , θ : T ac → RT . We refer to p and θ as

optimal if they solve (2.4), i.e., globally minimize the objective among
all feasible p and θ. We refer to p and θ as locally optimal if they are a
locally optimal point for (2.4).

Dual variables and locational marginal prices. Suppose p0 is a set of

optimal power schedules, that also minimizes the Lagrangian
T
(yp0 )t (τ )p̄t (τ ),
XX
f (p, θ) +
t∈T τ =1

subject to θ̃ = 0, where yp0 : T → RT are the dual variables associ-

ated with the power balance constraint p̄ = 0. (This is actually the
partial Lagrangian as we only dualize the power balance constraints,
but not the phase consistency constraints.) In this case we call yp0 a set
of optimal Lagrange multipliers or dual variables. When p0 is a locally
optimal point, which also locally minimizes the Lagrangian, then we
refer to yp0 as a set of locally optimal Lagrange multipliers.
The dual variables yp0 are related to the traditional concept of lo-
cational marginal prices L0 : T → RT by rescaling the dual variables
associated with each terminal according to the size of its associated
net, i.e., L0t = |n|(yp0 )t , where t ∈ n. This rescaling is due to the fact
that locational marginal prices are the dual variables associated with
the constraints in (2.1) rather than their scaled form used in (2.4) [18].

2.3 Discussion
We now describe our model in a less formal manner. Generators, loads,
energy storage systems, and other power sources and sinks are modeled
as single terminal devices. Transmission lines (or more generally, any
80 Network Model

wire or set of wires that conveys power), AC-DC converters, and AC

phase shifters are modeled as two-terminal devices. Terminals are ports
on a device through which power flows, either into or out of the device
(or both, at different times, as happens in a storage device). The flow
for AC terminals could be, e.g., three phase, two phase, single phase,
230V, or 230kV; the flow for DC terminals could be high voltage DC,
12V DC, or a floating voltage (e.g., the output of a solar panel). We
model these real cases with a different type for each mechanism (e.g.,
two and three phase AC terminals would have distinct types and could
not be connected to the same net).
Nets are used to model ideal lossless uncapacitated connections be-
tween terminals over which power is transmitted and physical con-
straints hold (e.g., equal voltages, currents summing to zero); losses,
capacities, and more general connection constraints between a set of
terminals can be modeled with the addition of a device and individual
nets which connect each terminal to the new device. An AC net cor-
responds to a direct connection between its associated terminals (e.g.,
a bus); all the terminals’ voltage phases must be the same and their
power is conserved. A two terminal DC net is just a wired connection
of the terminals. A DC net with more than two terminals is a smart
power router, which actively chooses how to distribute the incoming
and outgoing power flows among its terminals.
The objective function of a device is used to measure the cost (which
can be negative, representing revenue) associated with a particular
mode of operation, such as a given level of consumption or genera-
tion of power. This cost can include the actual direct cost of operating
according to the given power schedules, such as a fuel cost, as well as
other costs such as CO2 generation, or costs associated with increased
maintenance and decreased system lifetime due to structural fatigue.
The objective function can also include local variables other than power
and phase schedules, such as the state of charge of a storage device.
Constraints on the power and phase schedules and internal variables
for a device are encoded by setting the objective function to +∞ for
power and phase schedules that violate the constraints. In many cases,
a device’s objective function will only take on the values 0 and +∞,
2.4. Example 81

Figure 2.1: A simple network (left); its transformation into standard form (right).

indicating no local preference among feasible power and phase sched-

ules. Many devices, especially single-terminal devices such as loads or
generators, impose no constraints on their AC terminals’ phase angles;
in other words, these terminals have ‘floating’ voltage phase angles,
which are free to take any value.

2.4 Example
We illustrate how a traditional power network can be recast into our
network model in Figure 2.1. The original power network, shown on
the left, contains 2 loads, 3 buses, 3 transmission lines, 2 generators,
and a single battery storage system. We can transform this small power
grid into our model by representing it as a network with 11 terminals,
8 devices, and 3 nets, shown on the right of figure 2.1. Terminals are
shown as small filled circles. Single terminal devices, which are used
to model loads, generators, and the battery, are shown as boxes. The
transmission lines are two terminal devices represented by solid lines.
The nets are shown as dashed rounded boxes. Terminals are associated
with the device they touch and the net in which they are contained.
The set of terminals can be partitioned by either the devices they
are associated with, or the nets in which they are contained. Figure 2.2
shows the network in Figure 2.1 as a bipartite graph, with devices on
the left and nets on the right. In this representation, terminals are
represented by the edges of the graph.
82 Network Model

L1

G1

T1

G2

T2

L2

T3

Figure 2.2: The network in Figure 2.1 represented as a bipartite graph. Devices
(boxes) are shown on the left with their associated terminals (dots). The terminals
are connected to their corresponding nets (solid boxes) on the right.
 3
Device Examples

In this chapter we present several examples of how common devices can

be modeled in our framework. These examples are intentionally kept
simple, but could easily be extended with more refined objectives and
constraints. In these examples, it is easier to discuss operational costs
and constraints for each device separately. A device’s objective function
is equal to the device’s cost function unless any constraint is violated,
in which case we set the objective value to +∞. For all single terminal
devices, we describe their objective and constraints in the case of a DC
terminal. For AC terminal versions of one terminal devices, the cost
functions and constraints are identical to the DC case, and the device
imposes no constraints on the phase schedule.

3.1 Generators
A generator is a single-terminal device with power schedule pgen , which
generates power over a range, P min ≤ −pgen ≤ P max , and has ramp-
rate constraints
Rmin ≤ −Dpgen ≤ Rmax ,
which limit the change of power levels from one period to the next.
Here, the operator D ∈ R(T −1)×T is the forward difference operator,

83
84 Device Examples

defined as

(Dx)(τ ) = x(τ + 1) − x(τ ), τ = 1, . . . , T − 1.

The cost function for a generator has the separable form

T
X
ψgen (pgen ) = φgen (−pgen (τ )),
τ =1

where φ : R → R gives the cost of operating the generator at a given

power level over a single time period. This function is typically, but
not always, convex and increasing. It could be piecewise linear, or, for
example, quadratic:
φgen (x) = αx2 + βx,
where α, β > 0.
More sophisticated models of generators allow for them to be
switched on or off, with an associated cost each time they are turned on
or off. When switched on, the generator operates as described above.
When the generator is turned off, it generates no power but can still
incur costs for other activities such as idling.

3.2 Transmission lines

DC transmission line. A DC transmission line is a device with two
DC terminals with power schedules p1 and p2 that transports power
across some distance. The line has zero cost function, but the power
flows are constrained. The sum p1 +p2 represents the loss in the line and
is always nonnegative. The difference p1 −p2 can be interpreted as twice
the power flow from terminal one to terminal two. A DC transmission
line has a maximum flow capacity, given by
|p1 − p2 |
≤ C max ,
2
and the constraint
p1 + p2 − ℓ(p1 , p2 ) = 0,
where ℓ(p1 , p2 ) : RT × RT → RT+ is a loss function.
3.2. Transmission lines 85

For a simple model of the line as a series resistance R with average

terminal voltage V , we have [4]
R p1 − p2 2
 
ℓ(p1 , p2 ) = .
V2 2
A more sophisticated model for the capacity of a DC transmission
line includes a dynamic thermal model for the temperature of the line,
which (indirectly and dynamically) sets the maximum capacity of the
line. A simple model for the temperature at time τ , denoted ξ(τ ), is
given by the first order linear dynamics
ξ(τ + 1) = αξ(τ ) + (1 − α)ξ amb (τ ) + β(p1 (τ ) + p2 (τ )),
for τ = 1, . . . , T − 1, where ξ amb (τ ) is the ambient temperature at
time τ , and α and β are model parameters that depend on the thermal
properties of the line. The capacity is then dynamically modulated by
requiring that ξ ≤ ξ max , where ξ max is the maximum safe temperature
for the line.

AC transmission line. An AC transmission line is a device with two

AC terminals, with (real) power schedules p1 and p2 and terminal volt-
age phase angles θ1 and θ2 , that transmits power across some distance.
It has zero cost function, but the power flows and voltage phase angles
are constrained. Like a DC transmission line, the sum p1 + p2 repre-
sents the loss in the line, ℓ, and is always nonnegative. The difference
p1 − p2 can be interpreted as twice the power flow from terminal one
to terminal two. An AC line has a maximum flow capacity given by
|p1 − p2 |
≤ C max .
2
(A line temperature based capacity constraint as described for a DC
transmission line can also be used for AC transmission lines)
We assume the line is characterized by its (series) admittance g +ib,
with g > 0. (We consider the series admittance model for simplicity; for
a more general Π model, similar but more complicated equations can be
derived.) Under the common assumption that the voltage magnitude
is fixed at V [4], the power and phase schedules satisfy the relations
p1 − p2
p1 + p2 = 2gV 2 (1 − cos(θ2 − θ1 )), = bV 2 sin(θ2 − θ1 ),
2
86 Device Examples

which can be combined to give the relations

1 g
p1 + p2 = 2
(p1 + p2 )2 + 2 2 (p1 − p2 )2 ,
4gV 4b V
p1 − p2
= bV 2 sin(θ2 − θ1 ).
2
Transmission lines are rarely operated with a phase angle difference
exceeding 15◦ , and in this regime, the approximation sin(θ2 − θ1 ) ≈
θ2 − θ1 holds within 1%. This approximation, known as the ‘DC-OPF
approximation’ [4], is frequently used in power systems analysis and
transforms the second relation above into the relation
p1 − p2
= bV 2 (θ2 − θ1 ). (3.1)
2
Note that the capacity limit constrains |p1 − p2 |, which in turn con-
strains the phase angle difference |θ2 − θ1 |; thus we can guarantee that
our small angle approximation is good by imposing a capacity con-
straint (which is possibly smaller than the true line capacity).

3.3 Converters and interface devices

Inverter. An inverter is a device with a DC terminal, dc, and an
AC terminal, ac, that transforms power from DC to AC and has no
cost function. An inverter has a maximum power output C max and a
conversion efficiency κ ∈ (0, 1]. It can be represented by the constraints

−pac (τ ) = κpdc (τ ), 0 ≤ −pac (τ ) ≤ C max , τ = 1, . . . , T.

The voltage phase angle on the AC terminal, θac , is unconstrained.

Rectifier. A rectifier is a device with an AC terminal, ac, and a DC

terminal, dc, that transforms power from AC to DC and has no cost
function. A rectifier has a maximum power output C max and a conver-
sion efficiency κ ∈ (0, 1]. It can be represented by the constraints

−pdc (τ ) = κpac (τ ), 0 ≤ −pdc (τ ) ≤ C max , τ = 1, . . . , T.

The voltage phase angle on the AC terminal, θac , is unconstrained.

3.4. Storage devices 87

Phase shifter. A phase shifter is a device with two AC terminals,

which is a lossless energy carrier that decouples their phase angles and
has zero cost function. A phase shifter enforces the power balance and
capacity limit constraints

p1 (τ ) + p2 (τ ) = 0, |p1 (τ )| ≤ C max , τ = 1, . . . , T.

If the phase shifter can only support power flow in one direction, say,
from terminal 1 to terminal 2, then in addition we have the inequalities
p1 (τ ) ≥ 0, τ = 1, . . . , T . The voltage phase angles θ1 and θ2 are un-
constrained. (Indeed, this what a phase shifter is meant to do.) When
there is no capacity constraint, i.e., C max = ∞, we can think of a phase
shifter as a special type of net for AC terminals that enforces power
balance, but not voltage phase consistency. (However, we model it as
a device, not a net.)

External tie with transaction cost. An external tie is a connec-

tion to an external source of power. We represent this as a single
terminal device with power schedule pex . In this case, pex (τ )− =
max{−pex (τ ), 0} is the amount of energy pulled from the source, and
pex (τ )+ = max{pex (τ ), 0} is the amount of energy delivered to the
source, at time τ . We have the constraint |pex (τ )| ≤ E max (τ ), where
E max ∈ RT is the transaction limit.
We suppose that the prices for buying and selling energy are given
by c ± γ respectively, where c(τ ) is the midpoint price, and γ(τ ) > 0
is the difference between the price for buying and selling (i.e., the
transaction cost). The cost function is then

−(c − γ)T (pex )+ + (c + γ)T (pex )− = −cT pex + γ T |pex |,

where |pex |, (pex )+ , and (pex )− are all interpreted elementwise.

3.4 Storage devices

A battery is a single terminal energy storage device with power schedule
pbat , which can take in or deliver energy, depending on whether it is
charging or discharging. The charging and discharging rates are limited
by the constraints −Dmax ≤ pbat ≤ C max , where C max ∈ RT and
88 Device Examples

Dmax ∈ RT are the maximum charging and discharging rates. At time

τ , the charge level of the battery is given by local variables
τ
q(τ ) = q init +
X
pbat (t), τ = 1, . . . , T,
t=1

where q init is the initial charge. It has zero cost function and the charge
level must not exceed the battery capacity, i.e., 0 ≤ q(τ ) ≤ Qmax ,
τ = 1, . . . , T . It is common to constrain the terminal battery charge
q(T ) to be some specified value or to match the initial charge q init .
More sophisticated battery models include (possibly state-
dependent) charging and discharging inefficiencies as well as charge
leakage [26]. In addition, they can include costs which penalize exces-
sive charge-discharge cycling.
The same general form can be used to model other types of energy
storage systems, such as those based on super-capacitors, flywheels,
pumped hydro, or compressed air, to name just a few.

3.5 Loads
Fixed load. A fixed energy load is a single terminal device with zero
cost function which consists of a desired consumption profile, l ∈ RT .
This consumption profile must be satisfied in each period, i.e., we have
the constraint pload = l.

Thermal load. A thermal load is a single terminal device with power

schedule ptherm which consists of a heat store (room, cooled water reser-
voir, refrigerator), with temperature profile ξ ∈ RT , which must be
kept within minimum and maximum temperature limits, ξ min ∈ RT
and ξ max ∈ RT . The temperature of the heat store evolves as

ξ(τ + 1) = ξ(τ ) + (µ/c)(ξ amb (τ ) − ξ(τ )) − (η/c)ptherm (τ ),

for τ = 1, . . . , T − 1 and ξ(1) = ξ init , where 0 ≤ ptherm ≤ H max is the

cooling power consumption profile, H max ∈ RT is the maximum cooling
power, µ is the ambient conduction coefficient, η is the heating/cooling
efficiency, c is the heat capacity of the heat store, ξ amb ∈ RT is the
3.5. Loads 89

ambient temperature profile, and ξ init is the initial temperature of the

heat store. A thermal load has zero cost function.
More sophisticated models [27] include temperature-dependent
cooling and heating efficiencies for heat pumps, more complex dynamics
of the system whose temperature is being controlled, and additional ad-
ditive terms in the thermal dynamics, to represent occupancy or other
heat sources.

Deferrable load. A deferrable load is a single terminal device with

zero cost function that must consume a minimum amount of power
over a given interval of time, which is characterized by the constraint
PD
τ =A pload (τ ) ≥ E, where E is the minimum total consumption for
the time interval τ = A, . . . , D. The energy consumption in each time
period is constrained by 0 ≤ pload ≤ Lmax . In some cases, the load can
only be turned on or off in each time period, i.e., pload (τ ) ∈ {0, Lmax }
for τ = A, . . . , D.

Curtailable load. A curtailable load is a single terminal device which

does not impose hard constraints on its power requirements, but in-
stead penalizes the shortfall between a desired load profile l ∈ RT and
delivered power. In the case of a linear penalty, its cost function is

α1T (l − pload )+ ,

where (z)+ = max(0, z), pload ∈ RT is the amount of electricity deliv-

ered to the device, α > 0 is a penalty parameter, and 1 is the vector
with all components one. Extensions include time-varying and nonlin-
ear penalties on the energy shortfall.

Electric vehicle. An electric vehicle charging system is an example of

a device that combines aspects of a deferable load and a storage device.
We model it as a single terminal device with power schedule pev which
has a desired charging profile cdes ∈ RT and can be charged within
a time interval τ = A, . . . , D. To avoid excessive charge cycling, we
assume that the electric vehicle battery cannot be discharged back into
the grid (in more sophisticated vehicle-to-grid models, this assumption
90 Device Examples

is relaxed), so we have the constraints 0 ≤ pev ≤ C max , where C max ∈

RT is the maximum charging rate. We assume that cdes (τ ) = 0 for
τ = 1, . . . A − 1, cdes (τ ) = cdes (D) for τ = D + 1, . . . , T , and that the
charge level is given by
τ
q(τ ) = q init +
X
pev (t),
t=A

where q init is the initial charge when it is plugged in at time τ = A.

We can model electric vehicle charging as a deferrable load, where
we require a given charge level to be achieved at some time. A more
realistic model is as a combination of a deferrable and curtailable load,
with cost function
D
(cdes (τ ) − q(τ ))+ ,
X
α
τ =A

where α > 0 is a penalty parameter. Here cdes (τ ) is the desired charge

level at time τ , and cdes (τ ) − q(τ ))+ is the shortfall.
 4
Convexity

In this chapter we discuss the important issue of convexity, both of

devices and the resulting dynamic power flow problem.

4.1 Devices
We call a device convex if its objective function is convex. A network is
convex if all of its devices are convex. For convex networks, the D-OPF
is a convex optimization problem, which means that in principle we can
efficiently find a global solution [7]. When the network is not convex,
even finding a feasible solution for the D-OPF can become difficult,
and finding and certifying a globally optimal solution to the D-OPF
is generally intractable. However, special structure in many practical
power distribution problems can allow us to guarantee optimality.
In the examples from Chapter 3, the inverter, rectifier, phase shifter,
battery, fixed load, thermal load, curtailable load, electric vehicle, and
external tie are all convex devices using the constraints and objective
functions given. A deferrable load is convex if we drop the constraint
that it can only be turned on or off. We discuss the convexity properties
of the generator and AC and DC transmission lines next.

91
92 Convexity

4.2 Relaxations
One technique to deal with non-convex networks is to use convex re-
laxations. We use the notation g env to denote the convex envelope [52]
of the function g. There are many equivalent definitions for the con-
vex envelope, for example, g env = (g ∗ )∗ , where g ∗ denotes the convex
conjugate of the function g. We can equivalently define g env to be the
largest convex lower bound of g. If g is a convex, closed, proper (CCP)
function, then g = g env .
The relaxed dynamic optimal power flow problem (RD-OPF) is

minimize f env (p, θ)

(4.1)
subject to p̄ = 0, θ̃ = 0,

with variables p : T → RT , θ : T ac → RT . This is a convex opti-

mization problem, whose optimal value can in principle be computed
efficiently, and whose optimal objective value is a lower bound for the
optimal objective value of the D-OPF. In some cases, we can guarantee
a priori that a solution to the RD-OPF will also be a solution to the
D-OPF [56, 55] based on a property of the network objective such as
monotonicity or unimodularity. Even when the relaxed solution does
not satisfy all of the constraints in the unrelaxed problem, it can be
used as a starting point to help construct good, local solutions to the
unrelaxed problem. The suboptimality of these local solutions can then
be bounded by the gap between their network objective and the lower
bound provided by the solution to the RD-OPF. If this gap is small for
a given local solution, we can guarantee that it is nearly optimal.

Generator. When a generator is modeled as in Chapter 3 and is al-

ways powered on, it is a convex device. However, when given the ability
to be powered on and off, the generator is no longer convex. In this
case, we can relax the generator objective function so that its cost for
power production in each time period, given in Figure 4.1, is a convex
function. This allows the generator to produce power in the interval
[0, P min ].
4.2. Relaxations 93

1.6 1.6
1.4 1.4

gen (pgen )
φgen (pgen )

1.2 1.2
1.0 1.0

φenv
0.8 0.8
0.6 0.6
0.4 0.4
0 1 2 3 4 0 1 2 3 4
−pgen −pgen

Figure 4.1: Left: Cost function for a generator that can be turned off. Right: Its
convex relaxation.

p2 p2

p1 p1

Figure 4.2: Left: Feasible sets of a transmission lines with no loss (black) and AC
loss (grey). Right: Their convex relaxations.
94 Convexity

AC and DC transmission lines. In a lossless transmission line (AC or

DC), we have ℓ(p1 , p2 ) = 0, and thus the set of feasible power schedules
is the line segment

L = {(p1 , p2 ) | p1 = −p2 , p2 ∈ [−C max /2, C max /2]},

as shown in Figure 4.2 in black. When the transmission line has losses,
in most cases the loss function ℓ is a convex function of the input and
output powers, which leads to a feasible power region like the grey arc
in the left part of Figure 4.2.
The feasible set of a relaxed transmission line is given by the convex
hull of the original transmission line’s constraints. The right side of
figure 4.2 shows examples of this for both lossless and lossy transmission
lines. Physically, this relaxation gives lossy transmission lines the ability
to discard some additional power beyond what is simply lost to heat.
Since electricity is generally a valuable commodity in power networks,
the transmission lines will generally not throw away any additional
power in the optimal solution to the RD-OPF, leading to the power
line constraints in the RD-OPF being tight and thus also satisfying
the unrelaxed power line constraints in the original D-OPF. As was
shown in [40], when the network is a tree, this relaxation is always
tight. In addition, when all locational marginal prices are positive and
no other non-convexities exist in the network, the tightness of the line
constraints in the RD-OPF can be guaranteed in the case of networks
that have separate phase shifters on each loop in the networks whose
shift parameter can be freely chosen [54].
 5
Proximal Message Passing

In this chapter we describe our method for solving D-OPF. We begin by

deriving the proximal message passing algorithms assuming that all the
device objective functions are convex closed proper (CCP) functions.
We then compare the computational and communication requirements
of proximal message passing with a centralized solver for the D-OPF.
The additional requirements that the functions are closed and proper
are technical conditions that are in practice satisfied by any convex
function used to model devices. We note that we do not require either
finiteness or strict convexity of any device objective function, and that
all results apply to networks with arbitrary topologies.

Notation
Whenever we have a set of variables that maps terminals to time peri-
ods, x : T → RT (which we can also associate with a |T | × T matrix),
we will use the same index, over-line, and tilde notation for the variables
x as we do for power schedules p and phase schedules θ. For example,
xt ∈ RT consists of the time period vector of values of x associated
with terminal t, x̄t = (1/|n|) t′ ∈n xt′ , where t ∈ n, and x̃t = xt − x̄t ,
P

with similar notation for indexing x by devices and nets.

95
96 Proximal Message Passing

5.1 Derivation
We derive the proximal message passing equations by reformulating the
D-OPF using the alternating direction method of multipliers (ADMM)
and then simplifying the resulting equations. We refer the reader to [6]
for a thorough overview of ADMM.
We first rewrite the D-OPF as

minimize d∈D fd (pd , θd ) + n∈N (gn (zn ) + hn (ξn ))

P P
(5.1)
subject to p = z, θ = ξ

with variables p, z : T → RT , and θ, ξ : T ac → RT , where gn (zn ) is the

indicator function on the set {zn | z̄n = 0} and hn (ξn ) is the indicator
function on the set {ξn | ξ˜n = 0}. We use the notation from [6] and,
ignoring a constant, form the augmented Lagrangian
X X
Lρ (p, θ, z, ξ, u, v) = fd (pd , θd ) + (gn (zn ) + hn (ξn ))
d∈D n∈N
 
+ (ρ/2) kp − z + uk22 + kθ − ξ + vk22 , (5.2)

with the scaled dual variables u = yp /ρ : T → RT and v = yθ /ρ :

T ac → RT , which we associate with |T | × T and |T ac | × T matrices,
respectively, where yp : T → RT are the dual variables associated with
the power balance constraints and yθ : T ac → RT are the dual variables
associated with the phase consistency constraints. Because devices and
nets are each partitions of the terminals, the last two terms of (5.2)
can be split across either devices or nets, i.e.,

kp − z + uk22 = kpd − zd + ud k22 = kpn − zn + un k22 ,

X X

d∈D n∈N

vk22 vd k22 kθn − ξn + vn k22 .

X X
kθ − ξ + = kθd − ξd + =
d∈D n∈N

The resulting ADMM algorithm is then given by the iterations


k 2
(pk+1 k+1 k
d , θd ) := argmin fd (pd , θd ) + (ρ/2)(kpd − zd + ud k2
pd ,θd

+kθd − ξdk + vdk k22 ) , d ∈ D,
5.1. Derivation 97
 
2
znk+1 := argmin gn (zn ) + (ρ/2)kzn − ukn − pk+1
n k2 , n ∈ N,
zn
 
ξnk+1 := argmin hn (ξn ) + (ρ/2)kξn − vnk − θnk+1 k22 , n ∈ N,
ξn

uk+1
n := ukn + (pk+1
n − znk+1 ), n ∈ N,
vnk+1 := vnk + (θnk+1 − ξnk+1 ), n ∈ N,

where the first step is carried out in parallel by all devices, and then
the second and third and then fourth and fifth steps are carried out in
parallel by all nets.
Since gn (zn ) and hn (ξn ) are simply indicator functions for each
net n, the second and third steps of the algorithm can be computed
analytically and are given by

znk+1 := ukn + pk+1

n − ūkn − p̄k+1
n ,
ξnk+1 := v̄nk + θ̄nk+1 ,

respectively. From the second expression, it is clear that ξnk is simply

|n| copies of the same vector for all k.
Substituting these expressions into the u and v updates, the algo-
rithm can be simplified further to yield proximal message passing:

1. Prox schedule updates.

(pk+1 k+1 k k k k k−1

d , θd ) := proxfd ,ρ (pd − p̄d − ud , θ̄d + v̄d − vdk ), d ∈ D.

2. Scaled price updates.

uk+1
n := ukn + p̄k+1
n , n ∈ N,
vnk+1 := ṽnk + θ̃nk+1 , n ∈ N,

where the prox function for a function g is given by

 
proxg,ρ (x) = argmin g(y) + (ρ/2)kx − yk22 , (5.3)
y

and is guaranteed to exist when g is CCP [52]. If in addition v̄ 0 = 0

(note that any optimal dual variables v ⋆ must also satisfy v̄ ⋆ = 0), then
the v update simplifies to vnk+1 := vnk + θ̃nk+1 and the second argument
of the prox function in the first step simplifies to θ̄dk − vdk .
98 Proximal Message Passing

The origin of the name ‘proximal message passing’ should now be

clear. In each iteration, every device computes the prox function of its
objective function, with an argument that depends on messages passed
to it through its terminals by its neighboring nets in the previous it-
eration (p̄kd , θ̃dk , ukd , and vdk ). Then, every device passes to its terminals
the newly computed power and phase schedules, pk+1 d and θdk+1 , which
are then passed to the terminals’ associated nets. Every net computes
the prox function of the power balance and phase consistency indicator
functions (which corresponds to projecting the power and phase sched-
ules back to feasibility), computes the new average power imbalance,
p̄k+1
n and phase residual, θ̃nk+1 , updates its dual variables, uk+1 n and
k+1
vn , and broadcasts these values to its associated terminals’ devices.
Since p̄kn is simply |n| copies of the same vector for all k, all terminals
connected to the same net must have the same value for their dual vari-
ables associated with power balance throughout the algorithm, i.e., for
all values of k, ukt = ukt′ whenever t, t′ ∈ n for any n ∈ N .
As an example, consider the network represented by figures 2.1 and
2.2. The proximal message passing algorithm performs the power and
phase schedule updates on the devices (the boxes on the left in Fig-
ure 2.2). The devices share the respective power and phase profiles via
the terminals, and the nets (the solid boxes on the right) compute the
scaled price updates. For any network, the proximal message passing
algorithm can be thought of as alternating between devices (on the
left) and nets (on the right).

5.2 Convergence
Theory. We now comment on the convergence of proximal message
passing. Since proximal message passing is a version of ADMM, all
convergence results that hold for ADMM also hold for proximal message
passing. In particular, when all devices have CCP objective functions
and a feasible solution to the D-OPF exists, the following hold:

1. Power balance and phase consistency are achieved: p̄k → 0 and

θ̃k → 0 as k → ∞.

2. Operation is optimal: k k → f ⋆ as k → ∞.
d∈D fd (pd , θd )
P
5.2. Convergence 99

3. Optimal prices are found: ρuk = ypk → yp⋆ and ρv k = yθk → yθ⋆ as
k → ∞.

Here f ⋆ is the optimal value for the D-OPF, and yp⋆ and yθ⋆ are optimal
dual variables for the power schedule and phase consistency constraints,
respectively. The proof of these results (in the more general setting)
can be found in [6]. As a result of the third condition, the optimal
locational marginal prices L⋆ can be found for each net n ∈ N by
setting L⋆n = |n|(yp⋆ )n .

Stopping criterion. Following [6], we can define primal and dual resid-
uals, which for proximal message passing simplify to

rk = (p̄k , θ̃k ) sk = ρ((pk − p̄k ) − (pk−1 − p̄k−1 ), θ̄k − θ̄k−1 ).

We give a simple interpretation of each residual. The primal residual is

simply the net power imbalance and phase inconsistency across all nets
in the network, which is the original measure of primal feasibility in the
D-OPF. The dual residual is equal to the difference between the current
and previous iterations of both the difference between power schedules
and their average net power as well as the average phase angle on each
net. The locational marginal price at each net is determined by the
deviation of all associated terminals’ power schedule from the average
power on that net. As the change in these deviations approaches zero,
the corresponding locational marginal prices converge to their optimal
values, and all phase angles are consistent across all AC nets.
We can define a simple criterion for terminating proximal message
passing when
krk k2 ≤ ǫpri , ksk k2 ≤ ǫdual ,

where ǫpri and ǫdual are, respectively, primal and dual tolerances. We
can normalize both of these quantities to network size by the relation
q
pri dual abs
ǫ =ǫ =ǫ |T |T ,

for some absolute tolerance ǫabs > 0.

100 Proximal Message Passing

Choosing a value of ρ. Numerous examples show that the value of

ρ can have a strong effect on the rate of convergence of ADMM and
proximal message passing. Many good methods for picking ρ in both
offline and online fashions are discussed in [6]. We note that unlike
other versions of ADMM, the scaling parameter ρ enters very simply
into the proximal equations and can thus be modified online without
incurring any additional computational penalties, such as having to re-
factorize a matrix. For devices whose objectives just encode constraints
(i.e., only take on the values 0 and +∞), the prox function reduces to
projection, and is independent of ρ.
We can modify the proximal message passing algorithm with the
addition of a third step

3. Parameter update and price rescaling.

ρk+1 := h(ρk , rk , sk ),
ρk k+1
uk+1 := u ,
ρk+1
ρk k+1
v k+1 := v ,
ρk+1

for some function h. We desire to pick an h such that the primal

and dual residuals are of similar size throughout the algorithm, i.e.,
ρk krk k2 ≈ ksk k2 for all k. To accomplish this task, we use a simple
proportional-derivative controller to update ρ, choosing h to be

h(ρk ) = ρk exp(λwk + µ(wk − wk−1 )),

where wk = ρk krk k2 /ksk k2 −1 and λ and µ are nonnegative parameters

chosen to control the rate of convergence. Typical values of λ and µ
are between 10−3 and 10−1 .
When ρ is updated in such a manner, convergence is sped up in
many examples, sometimes dramatically. Although it can be difficult
to prove convergence of the resulting algorithm, a standard trick is
to assume that ρ is changed in only a finite number of iterations, af-
ter which it is held constant for the remainder of the algorithm, thus
guaranteeing convergence.
5.3. Discussion 101

Non-convex case. When one or more of the device objective func-

tions is non-convex, we can no longer guarantee that proximal message
passing converges to the optimal value of the D-OPF or even that it
converges at all (i.e., reaches a fixed point). Prox functions for non-
convex devices must be carefully defined as the set of minimizers in
(5.3) is no longer necessarily a singleton. Even then, prox functions of
non-convex functions are often intractable to compute.
One solution to these issues is to use proximal message pass-
ing to solve the RD-OPF. It is easy to show that f env (p, θ) =
env
d∈D fd (pd , θd ). As a result, we can run proximal message passing
P

using the prox functions of the relaxed device objective functions. Since
fdenv is a CCP function for all d ∈ D, proximal message passing in this
case is guaranteed to converge to the optimal value of the RD-OPF
and yield the optimal relaxed locational marginal prices.

5.3 Discussion
To compute the proximal messages, devices and nets only require
knowledge of who their network neighbors are, the ability to send small
vectors of numbers to those neighbors in each iteration, and the abil-
ity to store small amounts of state information and efficiently compute
prox functions (devices) or projections (nets). As all communication is
local and peer-to-peer, proximal message passing supports the ad hoc
formation of power networks, such as micro grids, and is self-healing
and robust to device failure and unexpected network topology changes.
Due to recent advances in convex optimization [61, 46, 47], many
of the prox function calculations that devices must perform can be
very efficiently executed at millisecond or microsecond time-scales on
inexpensive, embedded processors [30]. Since all devices and all nets
can each perform their computations in parallel, the time to execute a
single, network wide proximal message passing iteration (ignoring com-
munication overhead) is equal to the sum of the maximum computation
time over all devices and the maximum computation time of all nets in
the network. As a result, the computation time per iteration is small
and essentially independent of the size of the network.
In contrast, solving the D-OPF in a centralized fashion requires
102 Proximal Message Passing

complete knowledge of the network topology, sufficient communication

bandwidth to centrally aggregate all devices objective function data,
and sufficient centralized computational resources to solve the result-
ing D-OPF. In large, real-world networks, such as the smart grid, all
three of these requirements are generally unattainable. Having accu-
rate and timely information on the global connectivity of all devices
is infeasible for all but the smallest of dynamic networks. Centrally
aggregating all device objective functions would require not only infea-
sible bandwidth and data storage requirements at the aggregation site,
but also the willingness of all devices to expose what could be private
and/or proprietary function parameters in their objective functions.
Finally, a centralized solution to the D-OPF requires solving an opti-
mization problem with Ω(|T |T ) variables, which leads to an identical
lower bound on the time scaling for a centralized solver, even if problem
structure is exploited. As a result, the centralized solver cannot scale
to solve the D-OPF on very large networks.
 6
Numerical Examples

In this chapter we illustrate the speed and scaling of proximal message

passing with a range of numerical examples. In the first two sections, we
describe how we generate network instances for our examples. We then
describe our implementation, showing how multithreading can exploit
problem parallelism and how proximal message passing would scale in a
fully peer-to-peer implementation. Lastly, we present our results, and
demonstrate how the number of iterations needed for convergence is
essentially independent of network size and also significantly decreases
when the algorithm is seeded with a reasonable warm-start.

6.1 Network topology

We generate a network instance by first picking the number of nets
N . We generate the nets’ locations√xi ∈ R2 , i = 1, . . . , N by drawing
them uniformly at random from [0, N ]2 . (These locations will be used
to determine network topology.) Next, we introduce transmission lines
into the network as follows. We first connect a transmission line between
all pairs of nets i and j independently and with probability

γ(i, j) = α min(1, d2 /kxi − xj k22 ).

103
104 Numerical Examples

In this way, when the distance between i and j is smaller than d, they
are connected with a fixed probability α > 0, and when they are located
farther than distance d apart, the probability decays as 1/kxi − xj k22 .
After this process, we add a transmission line between any isolated net
and its nearest neighbor. We then introduce transmission lines between
distinct connected components by selecting two connected components
uniformly at random and then selecting two nets, one inside each com-
ponent, uniformly at random and connecting them by a transmission
line. We continue this process until the network is connected.
For the examples we present, we chose parameter values d = 0.11
and α = 0.8 as the parameters for generating our network. This results
in networks with an average degree of 2.1. Using these parameters, we
generated networks with 30 to 100000 nets, which resulted in optimiza-
tion problems with approximately 10 thousand to 30 million variables.

6.2 Devices
After we generate the network topology described above, we randomly
attach a single (one-terminal) device to each net according to the dis-
tribution in table 6.1. We also allow the possibility that a net acts as
a distributor and has no device attached to it other than transmission
lines. About 10% of the transmission lines are DC transmission lines,
while the other are AC transmission lines. The models for each device
and line in the network are identical to the ones given in Chapter 3,
with model parameters chosen in a manner we describe below.
For simplicity, our examples only include networks with the devices
listed below. For all devices, the time horizon was chosen to be T = 96,
corresponding to 15 minute intervals for 24 hour schedules, with the
time period τ = 1 corresponding to midnight.

Generator. Generators have the quadratic cost functions given in

Chapter 3 and are divided into three types: small, medium, and large.
In each case, the generator provides some idling power, so we set
Pmin = 0.01. Small generators have the smallest maximum power out-
put, but the largest ramp rates, while large generators have the largest
maximum power output, but the slowest ramp rates. Medium genera-
6.2. Devices 105

Device Fraction
None 0.4
Generator 0.4
Curtailable load 0.1
Deferrable load 0.05
Battery 0.05

Table 6.1: Fraction of devices present in the generated networks.

P min P max Rmax α β

Large 0.01 50 3 0.001 0.1
Medium 0.01 20 5 0.005 0.2
Small 0.01 10 10 0.02 1

Table 6.2: Generator parameters.

tors lie in between. Large generators are generally more efficient than
small and medium generators which is reflected in their cost function by
having smaller values of α and β. Whenever a generator is placed into
a network, its type is selected uniformly at random, and its parameters
are taken from the appropriate row in table 6.2.

Battery. Parameters for a given instance of a battery are generated

by setting q init = 0 and selecting Qmax uniformly at random from the
interval [20, 50]. The charging and discharging rates are selected to be
equal (i.e., C max = Dmax ) and drawn uniformly at random from the
interval [5, 10].

Fixed load. The load profile for a fixed load instance is a sinusoid,

l(τ ) = c + a sin(2π(τ − φ0 )/T ), τ = 1, . . . , T,

with the amplitude a chosen uniformly at random from the interval

[0.5, 1], and the DC term c chosen so that c = a + u, where u is chosen
uniformly at random from the interval [0, 0.1], which ensures that the
load profile remains elementwise positive. The phase shift φ0 is chosen
106 Numerical Examples

uniformly at random from the interval [60, 72], ensuring that the load
profile peaks between the hours of 3pm and 6pm.

Deferrable load. For an instance of a deferrable load, we choose E

uniformly at random from the interval [5, 10]. The start time index A
is chosen uniformly at random from the discrete set {1, . . . , (T − 9)/2}.
The end time index D is then chosen uniformly at random over the set
{A + 9, . . . , T }, so that the minimum time window to satisfy the load
is 10 time periods (2.5 hours). We set the maximum power so that it
requires at least two time periods to satisfy the total energy constraint,
i.e., Lmax = 5E/(D − A + 1).

Curtailable loads. For an instance of a curtailable load, the desired

load l is constant over all time periods with a magnitude chosen uni-
formly at random from the interval [5, 15]. The penalty parameter α is
chosen uniformly at random from the interval [0.1, 0.2].

AC transmission line. For an instance of an AC line, we set the volt-

age magnitude equal to 1 and choose its remaining parameters by first
solving the D-OPF with lossless, uncapacitated lines. Using flow values
given by the solution to that problem, we set C max = max(30, 10F max )
for each line, where F max is equal to the maximum flow (from the
lossless solution) along that line over all periods.
We use the loss function for transmission lines with a series admit-
tance g+ib given by (3.1). We choose a maximum phase angle deviation
(in degrees) in the interval [1, 5] and a loss of 1 to 3 percent of C max
when transmitting power at maximum capacity. Once the maximum
phase angle and the loss are determined, g is chosen to provide the
desired loss when operating at maximum phase deviation, while b is
chosen so the line operates at maximum capacity when at maximum
phase deviation.

DC transmission line. DC transmission lines are handled just like AC

transmission lines. We set R = g/b, where g and b are chosen using the
procedure for the AC transmission line.
6.3. Serial multithreaded implementation 107

6.3 Serial multithreaded implementation

Our D-OPF solver is implemented in C++, with the core proximal
message passing equations occupying fewer than 25 lines of C++ (ex-
cluding problem setup and class specifications). The code is compiled
with gcc 4.7.2 on a 32-core, 2.2GHz Intel Xeon processor with 512GB
of RAM running the Ubuntu OS. The processor supports hyperthread-
ing, so we have access to 64 independent threads. We used the compiler
option -O3 to leverage full code optimization.
To approximate a fully distributed implementation, we use gcc’s
implementation of OpenMP (version 3.1) and multithreading to paral-
lelize the computation of the prox functions for the devices. We use 64
threads to solve each example network. Assuming perfect load balanc-
ing among the cores, this means that 64 prox functions are being evalu-
ated in parallel. Effectively, we evaluate the prox functions by stepping
serially through the devices in blocks of size 64. We do not parallelize
the computation of the dual updates over the nets since the overhead
of spawning threads dominates the vector operations themselves.
The prox functions for fixed loads and curtailable loads are separa-
ble over τ and can be computed analytically. For more complex devices,
such as a generator, battery, or deferrable load, we compute the prox
function using CVXGEN [46]. The prox function for a transmission line
is computed by projecting onto the convex hull of the line constraints.
For a given network, we solve the associated D-OPF with an abso-
lute tolerance ǫabs = 10−3 . This translates to three digits of accuracy in
the solution. The CVXGEN solvers used to evaluate the prox operators
for some devices have an absolute tolerance of 10−8 . We set ρ = 1.

6.4 Peer-to-peer implementation

We have not yet created a fully peer-to-peer, bulk synchronous parallel
[60, 45] implementation of proximal message passing, but have carefully
tracked solve times in our serial implementation in order to facilitate a
first order analysis of such a system. In a peer-to-peer implementation,
the prox schedule updates occur in parallel across all devices followed by
(scaled) price updates occurring in parallel across all nets. As previously
 108 Numerical Examples

101 101

kpk k2 / |T |T
10−2 10−2
|f k − f ⋆ |/f ⋆

p
10−3 10−3

10−5 10−5

10−7 10−7
0 250 500 750 1000 0 250 500 750 1000
iter k iter k

Figure 6.1: The relative suboptimality (left) and primal infeasibility (right) of proxi-
mal message passing on a network instance with N = 3000 nets (1 million variables).
The dashed line shows when the stopping criterion is satisfied.

mentioned, the computation time per iteration is thus the maximum

time, over all devices, to evaluate the prox function of their objective,
added to the maximum time across all nets to project their terminal
schedules back to feasibility and update their existing price vectors.
Since evaluating the prox function for some devices requires solving a
convex optimization problem, whereas the price updates only require a
small number of vector operations that can be performed as a handful of
SIMD instructions, the compute time for the price updates is negligible
in comparison to the prox schedule updates. The determining factor in
solve time, then, is in evaluating the prox functions for the schedule
updates. In our examples, the maximum time taken to evaluate any
prox function is 1 ms.

6.5 Results
We first consider a single example: a network instance with N = 3000 (1
million variables). Figure 6.1 shows that after fewer than 200 iterations
of proximal message passing, both the relative suboptimality as well as
the average net power imbalance and average phase inconsistency are
both less than 10−3 . The convergence rates for other network instances
over the range of sizes we simulated are similar.
In Figure 6.2, we present average timing results for solving the D-
OPF for a family of examples, using our serial implementation, with
6.5. Results 109

networks of size N = 30, 100, 300, 1000, 3000, 10000, 30000, and
100000. For each network size, we generated and solved 10 network in-
stances to compute average solve times and confidence intervals around
those averages. The times were modeled with a log-normal distribution.
For network instances with N = 100000 nets, the problem has over 30
million variables, which we solve serially using proximal message pass-
ing in 5 minutes on average. By fitting a line to the proximal message
passing runtimes, we find that our parallel implementation empirically
scales as O(N 0.996 ), i.e., solve time is linear in problem size.
For a peer-to-peer implementation, the runtime of proximal message
passing should be essentially constant, and in particular independent of
the size of the network. To solve a problem with N = 100000 nets (30
million variables) with approximately 200 iterations of our algorithm
then takes only 200 ms. In practice, the actual solve time would clearly
be dominated by network communication latencies and actual runtime
performance will be determined by how quickly and reliably packets can
be delivered [34]. As a result, in a true peer-to-peer implementation, a
negligible amount of time is actually spent on computation. However,
it goes without saying that many other issues must be addressed with a
peer-to-peer protocol, including handling network delays and security.
Figure 6.2 shows cold start runtimes for solving the D-OPF. If we
have good estimates of the power and phase schedules and dual vari-
ables for each terminal, we can use them to warm start our D-OPF
solver. To show the effect, we randomly convert 5% of the devices into
fixed loads and solve a specific instance with N = 3000 nets (1 million
variables). Let K cold to be the number of iterations needed to solve an
instance of this problem. We then uniformly scale the load profiles of
each device by separate and independent lognormal random variables.
The new profiles, ˆl, are obtained from the original profiles l via
ˆl = l exp(σX),

where X ∼ N (0, 1), and σ > 0 is given. Using the original solution
to warm start our solver, we solve the perturbed problem and report
the number of iterations K warm needed. Figure 6.3 shows the ratio
K warm /K cold as we vary σ, showing the significant savings possible
with warm-starting even under relatively large perturbations.
110 Numerical Examples

1000

100
time (seconds)

10

0.1

10 100 1000 10000 100000

N
Figure 6.2: Average execution times for a family of networks on 64 threads. Error
bars show 95% confidence bounds. The dotted line shows the least-squares fit to the
data, resulting in a scaling exponent of 0.996.

1.0

0.8
K warm /K cold

0.6

0.4

0.2

0
0.00 0.05 0.10 0.15 0.20
σ
Figure 6.3: Relative number of iterations needed to converge from a warm start for
various perturbations of load profiles compared to original number of iterations.
 7
Extensions

Here, we give some possible extensions of our model and method.

7.1 Closed-loop control

So far, we have considered only a static energy planning problem, where
each device on the network plans power and phase schedules extending
T steps into the future and then executes all T steps. This ‘open loop’
control can fail spectacularly, since it will not adjust its schedules in
response to external disturbances that were unknown at the original
time the schedules were computed.
To alleviate this problem, we propose the use of receding horizon
control (RHC) [43, 3, 47] for dynamic network operation. In RHC, at
each time step τ , we determine a plan of action over a fixed time horizon
T into the future by solving the D-OPF using proximal message pass-
ing. The first step of all schedules is then executed, at which point the
entire process is repeated, incorporating new measurements, external
data, and predictions that have become available.
RHC has been successfully applied in many areas, including chemi-
cal process control [50], supply chain management [11], stochastic con-
trol in economics and finance [24, 58], and energy storage system op-

111
112 Extensions

eration [37]. While RHC is in general not an optimal controller, it has

been shown to achieve good performance in many different domains.
RHC is ideally suited for use with proximal message passing. First,
when one time step is executed, we can warm start the next round of
proximal message passing with the T − 1 schedules and dual variables
that were computed, but not otherwise executed, for each device and
each net in the previous iteration of RHC. As was shown in the previ-
ous section, this can dramatically speed up computation and allow for
RHC to operate network-wide at fraction of a second rates. In addition,
RHC does not require any stochastic or formal model of future predic-
tion uncertainty. While statistical predictions can be used if they are
available, predictions from other sources, such as analysts or markets,
are just as simple to integrate into RHC, and are also much easier to
come by in many real-world scenarios.
Perhaps the most important synergy between proximal message
passing and RHC is that the predictions used by each device need
only concern that one device and do not need to include any estimates
concerning other devices. This allows for devices to each use their own
form of prediction without worrying about what other devices exist or
what form of prediction they are using (e.g., even if one generator uses
statistical predictions, other devices need not).
The reason for this is that proximal message passing integrates all
the device predictions into the final solution — just as they would
have been in a centralized solution — but does so through the power
and phase schedules and prices that are shared between neighboring
devices. In this way, for example, a generator only needs to estimate
its cost of producing power at different levels over the time horizon T .
It does not need to predict any demand itself, as those predictions are
passed to it in the form of the power schedule and price messages it
receives from its neighbors. Similarly, loads only need to forecast their
own future demand and utility of power in each time period. Loads do
not need to explicitly predict future prices, as those are the result of
running proximal message passing over the network.
7.2. Security constrained optimal power flow 113

7.2 Security constrained optimal power flow

In the SC-OPF problem, we determine a set of contingency plans for
devices connected on a power network, which tell us the power flows
and phase angles each device will operate at under nominal system
operation, as well as in a set of specified contingencies or scenarios. The
contingencies can correspond, say, to failure or degraded operation of
a transmission line or generator, or a substantial change in a load. In
each scenario the powers and phases must satisfy the network equations
(taking into account any failures for that scenario), and they are also
constrained in various ways across the scenarios. Generators and loads,
for example, might be constrained to not change their power generation
or consumption in any non-nominal scenario.
As a variation on this, we can allow such devices to modify their
powers from the nominal operation values, but only over some range
or for a set amount of time. The goal is to minimize a composite cost
function that includes the cost (and constraints) of nominal operation,
as well as those associated with operation in any of the scenarios. Prox-
imal message passing allows us to parallelize the computation of many
different (and coupled) scenarios across each device while maintaining
decentralized communication across the network.

7.3 Hierarchical models and virtualized devices

The power grid has a natural hierarchy, with generation and trans-
mission occurring at the highest level and residential consumption and
distribution occurring at the most granular. Proximal message passing
can be easily extended into hierarchical interactions by scheduling mes-
sages on different time scales and between systems at similar levels of
the hierarchy [10]. By aggregating multiple devices into a virtual de-
vice (which themselves may be further aggregated into another virtual
device), our framework naturally allows for the formation of composite
entities such as virtual power plants and demand response aggregators.
Let D ⊆ D be a group of devices that are aggregated into a virtual
device, which we will also refer to as ‘D’. We use the notation that
terminal t ∈ D if there exists a device d ∈ D such that t ∈ d. The set of
114 Extensions

terminals {t | t ∈ D} can be partitioned into two sets, those terminals

whose associated net’s terminals are all associated with a device which
is part of D, and those who are not. These two sets can be though
of as those terminal in D which are purely ‘internal’ to D, and those
terminals which are not, as shown in Figure 7.1. These two sets are
given by

Din = {t ∈ D | ∀t′ ∈ nt , t′ ∈ D},

Dout = {t ∈ D | ∃t′ ∈ nt , t′ 6∈ D},

respectively, where nt is defined to be the net associated with termi-

nal t (i.e., t ∈ nt ). We let (pin , θin ) and (pout , θout ) denote the power
and phase schedules associated with the terminals in the sets Din and
Dout , respectively. Since the power and phase schedules (pin , θin ) never
directly leave the virtual device, they can be considered as internal
variables for the virtual device.
The objective function of the virtual device, fD (pout , θout ), is given
by the solution to the following optimization problem

minimize d∈D fd (pd , θd )

P

subject to p̄in = 0, θ̃in = 0,

with variables pd and θd for d ∈ D. A sufficient condition for

fD (pout , θout ) being a convex function is that all of the virtual device’s
constituent devices’ objective functions are convex functions [7].
By recursively applying proximal message passing at each level of
the aggregation hierarchy, we can compute the objective functions for
each virtual device. This process can be continued down to the indi-
vidual device level, at which point the device must compute the prox
function for its own objective function as the base case.
These models allow for the computations necessary to operate a
smart grid network to be virtualized since the computations specific
to each device do not necessarily need to be carried out on the de-
vice itself, but can be computed elsewhere (e.g., the servers of a vir-
tual power plant, centrally by an independent system operator, . . . ),
and then transmitted to the device for execution. As a result, hier-
archical modeling allows one to smoothly interpolate from completely
7.4. Local stopping criteria and ρ updates 115

Figure 7.1: Left: A simple network with four devices and two nets. Right: A hierar-
chical representation with only 2 devices at the highest level. All terminals connected
to the left-most net are internal to the virtual device.

centralized operation of the grid (i.e., all objectives and constraints

are gathered in a single location and solved), to a completely decen-
tralized architecture where all communication is peer to peer. At all
scales, proximal message passing offers all decision making entities an
efficient method to compute optimal power and phase schedules for the
devices under their control, while maintaining privacy of their devices’
objective functions and constraints.

7.4 Local stopping criteria and ρ updates

The stopping criterion and ρ update method in Chapter 5 currently re-
quire global device coordination (via the global primal and dual residu-
als each iteration). These could be computed in a decentralized fashion
by gossip algorithms [53], but this could require many rounds of gos-
sip in between each iteration of proximal message passing, significantly
increasing runtime. We are investigating methods to let individual de-
vices or terminals independently choose both the stopping criterion and
different values of ρ based only on local information such as the primal
and dual residuals of a device and its neighbors.
For dynamic operation another approach is to run proximal message
passing continuously, with no stopping criteria. In this mode, devices
and nets would exchange messages with each other indefinitely and ex-
ecute the first step of their schedules at given times (i.e., gate closure),
at which point they shift their moving horizon forward one time step
and continue to exchange messages.
 8
Conclusion

We have presented a fully decentralized method for dynamic network

energy management based on message passing between devices. Prox-
imal message passing is simple and highly extensible, relying solely
on peer to peer communication between devices that exchange energy.
When the resulting network optimization problem is convex, proximal
message passing converges to the optimal value and gives optimal power
and phase schedules and locational marginal prices. We have presented
a parallel implementation that shows the time per iteration and the
number of iterations needed for convergence of proximal message pass-
ing are essentially independent of the size of the network. As a result,
proximal message passing can scale to extremely large networks with
almost no increase in solve time.

116
 Acknowledgments

The authors thank Yang Wang and Neal Parikh for extensive discus-
sions on the problem formulation as well as ADMM methods; Yang
Wang, Brendan O’Donoghue, Haizi Yu, Haitham Hindi, and Mikael
Johansson for discussions on optimal ρ selection and for help with the
ρ update method; Steven Low for discussions about end-point based
control; and Ed Cazalet, Ram Rajagopal, Ross Baldick, David Chas-
sin, Marija Ilic, Trudie Wang, and Jonathan Yedidia for many helpful
comments. We would like to thank Marija Ilic, Le Xie, and Boris De-
fourny for pointing us to DYNMONDS and other earlier Lagrangian
approaches. We are indebted to Misha Chertkov, whose questions on
an early version of this paper prodded us to make the concept of AC
and DC terminals explicit. Finally, we thank Warren Powell and Hugo
Simao for encouraging us to release implementations of these methods.
This research was supported in part by Precourt 1140458-
1-WPIAE, by AFOSR grant FA9550-09-1-0704, by NASA grant
NNX07AEIIA, and by the DARPA XDATA grant FA8750-12-2-0306.
After this paper was submitted, we became aware of [31] and [32],
which apply ADMM to power networks for the purpose of robust state
estimation. Our paper is independent of their efforts.

117
 References

[1] R. Baldick, Applied Optimization: Formulation and Algorithms for Engineering

Systems. Cambridge University Press, 2006.
[2] S. Barman, X. Liu, S. Draper, and B. Recht, “Decomposition methods for large
scale LP decoding,” Submitted, IEEE Transactions on Information Theory,
2012.
[3] A. Bemporad, “Model predictive control design: New trends and tools,” in
Proceedings of 45th IEEE Conference on Decision and Control, pp. 6678–6683,
2006.
[4] A. Bergen and V. Vittal, Power Systems Analysis. Prentice Hall, 1999.
[5] D. P. Bertsekas, Constrained Optimization and Lagrange Multiplier Methods.
Academic Press, 1982.
[6] S. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, “Distributed optimiza-
tion and statistical learning via the alternating direction method of multipliers,”
Foundations and Trends in Machine Learning, vol. 3, pp. 1–122, 2011.
[7] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge University
Press, 2004.
[8] J. Carpentier, “Contribution to the economic dispatch problem,” Bull. Soc.
Francaise Elect., vol. 3, no. 8, pp. 431–447, 1962.
[9] E. A. Chakrabortty and M. D. Ilic, Control & Optimization Methods for Electric
Smart Grids. Springer, 2012.
[10] M. Chiang, S. Low, A. Calderbank, and J. Doyle, “Layering as optimization
decomposition: A mathematical theory of network architectures,” Proc. of the
IEEE, vol. 95, no. 1, pp. 255–312, 2007.
[11] E. G. Cho, K. A. Thoney, T. J. Hodgson, and R. E. King, “Supply chain plan-
ning: Rolling horizon scheduling of multi-factory supply chains,” in Proceedings
of the 35th conference on Winter simulation: driving innovation, pp. 1409–1416,
2003.

118
References 119

[12] B. H. Chowdhury and S. Rahman, “A review of recent advances in economic

dispatch,” IEEE Transactions on Power Systems, vol. 5, no. 4, pp. 1248–1259,
1990.
[13] A. O. Converse, “Seasonal energy storage in a renewable energy system,” Pro-
ceedings of the IEEE, vol. 100, pp. 401–409, Feb 2012.
[14] G. B. Dantzig and P. Wolfe, “Decomposition principle for linear programs,”
Operations Research, vol. 8, pp. 101–111, 1960.
[15] J. Eckstein, “Parallel alternating direction multiplier decomposition of convex
programs,” Journal of Optimization Theory and Applications, vol. 80, no. 1,
pp. 39–62, 1994.
[16] J. H. Eto and R. J. Thomas, “Computational needs for the next generation
electric grid,” in Department of Energy, 2011. http://certs.lbl.gov/pdf/lbnl-
5105e.pdf.
[17] H. Everett, “Generalized Lagrange multiplier method for solving problems of
optimum allocation of resources,” Operations Research, vol. 11, no. 3, pp. 399–
417, 1963.
[18] A. Eydeland and K. Wolyniec, Energy and Power Risk Management: New De-
velopments in Modeling, Pricing and Hedging. Wiley, 2002.
[19] G. Forney, “Codes of graphs: Normal realizations,” IEEE Transactions on In-
formation Theory, vol. 47, no. 2, pp. 520–548, 2001.
[20] D. Gabay and B. Mercier, “A dual algorithm for the solution of nonlinear
variational problems via finite element approximations,” Computers and Math-
ematics with Applications, vol. 2, pp. 17–40, 1976.
[21] R. Glowinski and A. Marrocco, “Sur l’approximation, par elements finis d’ordre
un, et la resolution, par penalisation-dualité, d’une classe de problems de Dirich-
let non-lineares,” Revue Française d’Automatique, Informatique, et Recherche
Opérationelle, vol. 9, pp. 41–76, 1975.
[22] H. H. Happ, “Optimal power dispatch — a comprehensive survey,” IEEE Trans-
actions on Power Apparatus and Systems, vol. 96, no. 3, pp. 841–854, 1977.
[23] E. K. Hart, E. D. Stoutenburg, and M. Z. Jacobson, “The potential of intermit-
tent renewables to meet electric power demand: current methods and emerging
analytical techniques,” Proceedings of the IEEE, vol. 100, pp. 322–334, Feb
2012.
[24] F. Herzog, Strategic Portfolio Management for Long-Term Investments: An
Optimal Control Approach. PhD thesis, ETH, Zurich, 2005.
[25] M. R. Hestenes, “Multiplier and gradient methods,” Journal of Optimization
Theory and Applications, vol. 4, pp. 302–320, 1969.
[26] K. Heussen, S. Koch, A. Ulbig, and G. Andersson, “Energy storage in power
system operation: The power nodes modeling framework,” in Innovative Smart
Grid Technologies Conference Europe (ISGT Europe), 2010 IEEE PES, pp. 1–
8, Oct 2010.
[27] T. Hovgaard, L. Larsen, J. Jorgensen, and S. Boyd, “Nonconvex model pre-
dictive control for commercial refrigeration,” International Journal of Control,
vol. 86, no. 8, pp. 1349–1366, 2013.
120 References

[28] M. Ilic, L. Xie, and J.-Y. Joo, “Efficient coordination of wind power and price-
responsive demand — Part I: Theoretical foundations,” IEEE Transactions on
Power Systems, vol. 26, pp. 1875–1884, Nov 2011.
[29] M. Ilic, L. Xie, and J.-Y. Joo, “Efficient coordination of wind power and price-
responsive demand—part ii: Case studies,” IEEE Transactions on Power Sys-
tems, vol. 26, pp. 1885–1893, Nov 2011.
[30] J. L. Jerez, P. J. Goulart, S. Richter, G. A. Constantinides, E. C. Kerrigan,
and M. Morari, “Embedded online optimization for model predictive control at
megahertz rates,” Submitted, IEEE Transactions on Automatic Control, 2013.
[31] V. Kekatos and G. Giannakis, “Joint power system state estimation and breaker
status identification,” in Proceedings of the 44th North American Power Sym-
posium, 2012.
[32] V. Kekatos and G. Giannakis, “Distributed robust power system state estima-
tion,” IEEE Transactions on Power Systems, 2013.
[33] F. P. Kelly, A. K. Maulloo, and D. K. H. Tan, “Rate control in communication
networks: shadow prices, proportional fairness and stability,” Journal of the
Operational Research Society, vol. 49, pp. 237–252, 1998.
[34] A. Kiana and A. Annaswamy, “Wholesale energy market in a smart grid: A
discrete-time model and the impact of delays,” in Control and Optimization
Methods for Electric Smart Grids, (A. Chakrabortty and M. Ilic, eds.), pp. 87–
110, Springer US, 2012.
[35] B. H. Kim and R. Baldick, “Coarse-grained distributed optimal power flow,”
IEEE Transactions on Power Systems, vol. 12, no. 2, pp. 932–939, 1997.
[36] B. H. Kim and R. Baldick, “A comparison of distributed optimal power flow
algorithms,” IEEE Transactions on Power Systems, vol. 15, no. 2, pp. 599–604,
2000.
[37] M. Kraning, Y. Wang, E. Akuiyibo, and S. Boyd, “Operation and configuration
of a storage portfolio via convex optimization,” in Proceedings of the 18th IFAC
World Congress, pp. 10487–10492, 2011.
[38] A. Lam, B. Zhang, and D. Tse, “Distributed algorithms for optimal power flow
problem,” http://arxiv.org/abs/1109.5229, 2011.
[39] J. Lavaei and S. Low, “Zero duality gap in optimal power flow problem,” IEEE
Transactions on Power Systems, vol. 27, no. 1, pp. 92–107, 2012.
[40] J. Lavaei, D. Tse, and B. Zhang, “Geometry of power flows in tree networks,”
IEEE Power & Energy Society General Meeting, 2012.
[41] J. Liang, G. K. Venayagamoorthy, and R. G. Harley, “Wide-area measurement
based dynamic stochastic optimal power flow control for smart grids with high
variability and uncertainty,” IEEE Transactions on Smart Grid, vol. 3, pp. 59–
69, 2012.
[42] S. H. Low, L. Peterson, and L. Wang, “Understanding tcp vegas: a duality
model,” in Proceedings of the 2001 ACM SIGMETRICS international con-
ference on Measurement and modeling of computer systems, (New York, NY,
USA), pp. 226–235, ACM, 2001.
References 121

[43] J. Maciejowski, Predictive Control with Constraints. Prentice Hall, 2002.

[44] S. H. Madaeni, R. Sioshansi, and P. Denholm, “How thermal energy storage
enhances the economic viability of concentrating solar power,” Proceedings of
the IEEE, vol. 100, pp. 335–347, Feb 2012.
[45] G. Malewicz, M. H. Austern, A. J. C. Bik, J. C. Dehnert, I. Horn, N. Leiser, and
G. Czajkowski, “Pregel: A system for large-scale graph processing,” in Proceed-
ings of the 2010 International Conference on Management of Data, pp. 135–
146, 2010.
[46] J. Mattingley and S. Boyd, “CVXGEN: Automatic convex optimization code
generation,” http://cvxgen.com/, 2012.
[47] J. Mattingley, Y. Wang, and S. Boyd, “Receding horizon control: Automatic
generation of high-speed solvers,” IEEE Control Systems Magazine, vol. 31,
pp. 52–65, 2011.
[48] W. F. Pickard, “The history, present state, and future prospects of underground
pumped hydro for massive energy storage,” Proceedings of the IEEE, vol. 100,
pp. 473–483, Feb 2012.
[49] M. J. D. Powell, “A method for nonlinear constraints in minimization prob-
lems,” in Optimization, (R. Fletcher, ed.), Academic Press, 1969.
[50] S. J. Qin and T. A. Badgwell, “A survey of industrial model predictive control
technology,” Control Engineering Practice, vol. 11, no. 7, pp. 733–764, 2003.
[51] P. Ravikumar, A. Agarwal, and M. J. Wainwright, “Message-passing for graph-
structured linear programs: Proximal methods and rounding schemes,” Journal
of Machine Learning Research, vol. 11, pp. 1043–1080, 2010.
[52] R. T. Rockafellar, Convex Analysis. Princeton University Press, 1970.
[53] D. Shah, “Gossip algorithms,” Foundations and Trends in Networking, vol. 3,
no. 2, pp. 1–125, 2008.
[54] S. Sojoudi and J. Lavaei, “Physics of power networks makes hard problems easy
to solve,” To appear, IEEE Power & Energy Society General Meeting, 2012.
[55] S. Sojoudi and J. Lavaei, “Convexification of generalized network flow prob-
lem with application to power systems,” Preprint available at http://www.ee.
columbia.edu/~lavaei/Generalized_Net_Flow.pdf, 2013.
[56] S. Sojoudi and J. Lavaei, “Semidefinite relaxation for nonlinear optimization
over graphs with application to power systems,” Preprint available at http:
//www.ee.columbia.edu/~lavaei/Opt_Over_Graph.pdf, 2013.
[57] N. Taheri, R. Entriken, and Y. Ye, “A dynamic algorithm for facilitated charg-
ing of plug-in electric vehicles,” arxiv:1112:0697, 2011.
[58] K. T. Talluri and G. J. V. Ryzin, The Theory and Practice of Revenue Man-
agement. Springer, 2004.
[59] K. Turitsyn, P. Sulc, S. Backhaus, and M. Chertkov, “Options for control of re-
active power by distributed photovoltaic generators,” Proceedings of the IEEE,
vol. 99, pp. 1063–1073, 2011.
[60] L. G. Valiant, “A bridging model for parallel computation,” Communications
of the ACM, vol. 33, no. 8, p. 111, 1990.
122 References

[61] Y. Wang and S. Boyd, “Fast model predictive control using online optimiza-
tion,” IEEE Transactions on Control Systems Technology, vol. 18, pp. 267–278,
2010.
[62] J. Zhu, Optimization of Power System Operation. Wiley-IEEE Press, 2009.