Privacy-preserving Consensus-based Energy

Management in Smart Grid

Chengcheng Zhao Jianping He Peng Cheng
College of Control Department of Automation and Jiming Chen
Science and Engineering Shanghai Jiao Tong University China College of Control
Zhejiang University China Email: [email protected] Science and Engineering
Email:[email protected] Zhejiang University China
Email: {pcheng, jmchen}@iipc.zju.edu.cn

Abstract—This paper investigates the privacy-preserving prob- directed communication as well as the transmission losses
lem of the distributed consensus-based energy management into consideration, another consensus-based energy manage-
considering both generation units and responsive demands in ment algorithm was developed in [13]. However, all these
smart grid. First, we reveal the private information of consumers
including the electricity consumption and the sensitivity of the existing works neglect the privacy concern of the responsive
electricity consumption to the electricity price can be disclosed consumers. In [14], the disclosure of generation units’ private
without any privacy-preserving strategy. Then, we propose a information, i.e., the final generation power and the parameters
privacy-preserving algorithm to preserve the private information of cost functions in the consensus-based economic dispatch
of consumers through designing the secret functions, and adding process was analyzed, and then authors proposed a privacy-
zero-sum and exponentially decreasing noises. We also prove that
the proposed algorithm can preserve the privacy while keeping preserving strategy. But the proposed strategy is only effective
the optimality of the final state and the convergence performance for the fully undirected communication topology. Furthermore,
unchanged. Extensive simulations validate the theoretical results plentiful achievements for the privacy-preserving demand re-
and demonstrate the effectiveness of the proposed algorithm. sponse through the encryption or aggregation communication
design have been attained [15]. However, in the consensus-
I. I NTRODUCTION based DEM, these strategies can lead to much computation and
Furnished with smart meters, smart appliances and distribut- communication burden due to the asymptotically convergence.
ed energy resources, smart grid brings about the distributed Meanwhile, differential privacy has been considered as a
control and energy management [1]. Especially, Distributed significant method to preserve the privacy of the distributed
Energy Management (DEM) aims to improve the efficiency, optimization problem [16]. But it cannot guarantee that the
scalability and robustness of power grid through the distributed final solution is the exact optimal one. Therefore, how to de-
scheduling of both distributed energy resources and responsive sign the privacy-preserving consensus-based DEM algorithm
consumers by utilizing the information technology [2]. As a considering both of the generation and demand sides under
vitally important concern of consumers, the private informa- directed communications while guaranteeing the optimality of
tion such as habits and behaviors can be disclosed through the the final solution is still an open problem.
broadcast information during the participation in DEM. To deal with the above problem, we first provide the
Numerous endeavors have been devoted to designing consensus-based DEM algorithm considering both sides and
DEM algorithms in smart grid [3]–[12]. In [3]–[5], DEM directed communication topology. Then, we investigate how
problems with different goals are taken into consideration and what kind of the private information of consumers will be
and nonconsensus-based distributed algorithms are designed, disclosed under the algorithm. The privacy-preserving method
which mainly focus on the demand side management (DSM). is designed to conserve the private information through secret
Compared with these protocols, the consensus-based DEM functions and adding noises inspired by our previous work
algorithms can fulfill the global energy management target [17], under which the optimal solution can be achieved exactly.
only through local interactions. Hence, the consensus-based The main contributions of this paper are as follows.
DEM approaches are more flexible, scalable, robust and dis-
tributed. Much attention has been paid to the consensus-based 1. We investigate how the private information including the
economic dispatch problem of smart grid [6], [8]–[10], which electricity consumption and the sensitivity of the electricity
only considers the generation side management. In smart consumption to the electricity price will be disclosed under
grid, considering both of the generation and demand sides in the provided consensus-based algorithm.
DEM reveals the fairness between consumers and producers, 2. We propose the privacy-preserving consensus-based DEM
which makes the DEM problem more realistic and meaningful. algorithm (P-CEMA) by adding noises to the broadcast
Authors in [11], [12] proposed a consensus-based DEM algo- information to preserve the sensitivity and designing secret
rithm for two sides energy scheduling, which is only effective functions to conserve the electricity consumption.
for undirected connected communication network. Taking the 3. We prove that under P-CEMA, the private information

‹,(((
 can be preserved while the convergence and the optimal by adjusting Pj for a period due to the excitation of the
solution can be maintained. electricity price.
The remainder of this paper is organized as follows. Section Therefore, the social welfare, i.e., the sum of profits of all
II provides the models and formulations of the problem, and units and consumers, is expressed by,
   
the detailed privacy disclosure is presented in Section III. πig (Pi ) + πjd (Pj ) = −Ci (Pi ) + Uj (Pj ).
(1)
Section IV provides the privacy-preserving consensus-based i∈Vg j∈Vd i∈Vg j∈Vd

DEM algorithm and analyzes the performance. Section V tests We formulate the following optimization problem,
 
the main results through numerical examples and simulations. min Ci (Pi ) − Uj (Pj )
Conclusion is given in Section VI. i∈Vg j∈Vd
 
s.t. Pi = Pj (2a)
II. P RELIMINARIES AND P ROBLEM F ORMULATION i∈Vg j∈Vd

A. Network Model Pim ≤ Pi ≤ PiM , i ∈ V. (2b)

We consider a smart grid with ng generation units and where the constraint (2a) describes the balance of the genera-
nd responsive consumers indexed by 1, 2, · · · , ng and ng + tion and demand, and (2b) are the local power constraints for
1, · · · , ng + nd ≥ 3, respectively, where n = ng + nd . We units and consumers, respectively. It is assumed that the total
represent the communication network by a strongly connected generation capacity can always meet the demand requirement,
(any two nodes are connected by a directed path) directed  m  m  M  M
i.e., Pi ≤ Pj ≤ Pj ≤ Pi .
graph G = {V, E} with V being the set of n nodes and i∈Vg j∈Vd j∈Vd i∈Vg
E ⊂ V × V being the edge set. The unit and consumer
sets are expressed by Vg and Vd , respectively, satisfying C. Consensus-based Energy Management Algorithm
V = Vg ∪ Vd . Note that (j, i) ∈ E if and only if (iff ) A row stochastic matrix W = [wij ] and a column stochastic
matrix Q = [qij ] are introduced, i.e.,
node i can receive information from node j, i.e., node j
⎧ ⎧
is the in-neighbor of node i. The in-neighbor set of node ⎪
1
,j ∈ Ni+ ⎪
1
,j ∈ Ni+
⎪
⎪ |Ni+ |+1 ⎪
⎪ |Nj− |+1
i is denoted by Ni+ = {j|(j, i) ∈ E, j = i} and |Ni+ | is ⎨  ⎨ 
wij = 1− wij , j = i , qij = 1− qji , j =i .
the cardinality. Meanwhile, the out-neighbor set of node i is ⎪
⎪ j∈Ni+ ⎪
⎪ j∈Ni−
⎪
⎩ ⎪
⎩
Ni− = {j|(i, j) ∈ E, j = i} and the cardinality is |Ni− |. We 0, j / Ni+ , j = i
∈ 0, j / Ni+ , j =
∈ i
do not consider self loops in this paper, i.e., (i, i) ∈
/ E, ∀i ∈ V .
Then, we provide the following distributed algorithm to solve
B. Distributed Energy Management problem (2), namely, CEMA [13].
Social welfare maximization not only can maximize the The initialization of λi (0), Pi (0) and ξi (0) , ∀i ∈ V is given
total welfare of the whole society in electricity market, but also as below,   
can guarantee that each individual’s profit is maximized [18]. Ci Pim , i ∈ Vg ,
λi (0) =   , Pi (0) = ξi (0) = 0, i ∈ V (3)
Ui PiM , i ∈ Vd ,
Therefore, the objective of DEM is to maximize the total social
welfare by adjusting the electricity price while balancing the After initialization, each node broadcasts λi (0) and ξi (0). Af-
generation and demand. Here, we use price p as a variable to ter receiving all neighbors’ information, each node i executes
the following iteration process,
realize coordination among units and consumers. We consider 
one period DEM problem and introduce the profit functions λi (k + 1) = wij λj (k) + ηξi (k), i ∈ V (4a)
j∈V
of units and consumers, respectively. ⎧
Profit function of each generation unit: The generation ⎪
⎪ arg min [Ci (Pi (k)) − λi (k + 1)Pi (k)], i ∈ Vg
⎨ Pim ≤Pi (k)≤PiM
power of each unit i, i ∈ Vg is denoted by Pi with the Pi (k + 1) =
⎪ arg min [λ (k + 1)Pi (k) − Ui (Pi (k))], i ∈ Vd
⎪
⎩P m ≤P (k)≤P M i
lower (upper) bound expressed by Pim (PiM ). The cost of i i i
unit i can be represented by the quadratic function of Pi , ⎧

(4b)
i.e., Ci (Pi ) = ai Pi2 + bi Pi + ci , where ai , bi , ci are fitting ⎪
⎨ qij ξj (k) + Pi (k) − Pi (k + 1), i ∈ Vg
parameters. Then, we provide the profit πig (Pi ) of unit i as
j∈V
ξi (k + 1) =
(4c)
⎪
⎩ qij ξj (k) + Pi (k + 1) − Pi (k), i ∈ Vd .
πig (Pi ) = pPi − Ci (Pi ) = −ai Pi2 + (p − bi ) Pi − ci . j∈V

Profit function of each consumer: The utility function of where η is a constant satisfying 0 < η < 1.
each consumer j, j ∈ Vd is denoted by Uj (Pj ), which Private information of consumers: Here, we are concerned
characterizes the satisfaction degree of consumer j consuming about the private information of consumers, i.e., the quantity of
Pj . It is usually assumed that Uj (Pj ) satisfies the following the electricity consumption and the parameters of consumers’
three properties [19]: 1) The first order derivative Uj (Pj ) utility function which reveals the sensitivity of the electricity
satisfies Uj (Pj ) ≥ 0; 2) The second order derivative Uj (Pj ) consumption to the electricity price. By obtaining the contin-
satisfies Uj (Pj ) ≤ 0; 3) Uj (0) = 0. Then, the profit function uous electricity consumption of some consumer for multiple
of each consumer j is introduced as πjd (Pj ) = Uj (Pj ) − pPj . periods, the behaviors and habits of the consumer can be
For each responsive consumer j, it has a certain adjustable learned. Meanwhile, the sensitivity can be obtained and thus
range of power (from Pjm to PjM ) to maximize its own profit the future behaviors of the consumer can be predicted.

‹,(((
 Although under CEMA, the private information will not IV. P RIVACY- PRESERVING A LGORITHM D ESIGN
be broadcast directly, the private information can still be In this section, we first analyze whether there are noises, by
inferred through the broadcast infromation as the broadcast adding which to CEMA, we can guarantee the convergence
information is the function of the private information shown and optimality. Then, we propose a privacy-preserving proto-
in the iteration rule. Hence, it is desirable to preserve the col and analyze how the proposed protocol can preserve the
private information under CEMA. Since the iteration rule private information of consumers.
is based on the consensus, the privacy-preserving average
consensus [17] renders the privacy-preserving strategy design A. Noise Adding Analysis
for CEMA possible. However, different from the privacy- Here we provide the sufficient condition and the necessary
preserving average consensus, we need to preserve the final condition for CEMA with noises added to the broadcast
state and the functional relationship between variables under information λi (k) and ξi (k), ∀i ∈ V . We denote the noise
CEMA which is much more complex than average consensus. added to λi (k) and ξi (k) by ϑi (k) and θi (k), respectively.
The lemmas referred to [20] are as follows.
III. P RIVACY D ISCLOSURE
Lemma 1: If there exists a constant H ≥ such that
∞
 ∞

To analyze the privacy disclosure of CEMA, we first provide
|ϑi (k)| ≤ H, |θi (k)| ≤ H, i ∈ V, (7)
the sufficient and necessary condition for the initial state ξi (0) k=0 k=0
and Pi (0), which can guarantee the optimality of the final then CEMA achieves asymptotic convergence, i.e.,
solution. It can be proved by contradiction and the proof is lim λi (k) = λ̄, lim ξ( k) = 0, ∀i ∈ V. (8)
k→∞ k→∞
omitted here due to the space limitation. Lemma 2: If CEMA can achieve asymptotic convergence
Theorem 1: Suppose that the selected η can guarantee the and the final state is optimal, i.e.,
convergence of (4). If
and only if lim λi (k) = λ∗ , lim ξ( k) = 0, ∀i ∈ V, (9)
k→∞ k→∞
(ξi (0) + Pi (0)) = 0, (5) ∗
where λ is the optimal price, then there hold
i∈V

k 
the final solution is the optimal one of problem (2). lim ϑi (k) = lim θi (k) = 0, lim θi (j) = 0. (10)
k→∞ k→∞ k→∞
To make the initial states satisfy (5) without any control j=0 i∈V
operator, the simplest and direct method is to set ξi (0) = In fact, by adding decreasing and zero-sum noises to each
Pi (0) = 0, ∀i ∈ V or ξi (0) = −Pi (0), ∀i ∈ V . Under local power mismatch, we cannot preserve the privacy of the
this initialization way, as ξj (0) will be broadcast, Pj (0) will electricity consumption. By adding the noise θi (k) to ξi (k +1)
be known by the eavesdropper or the neighbors of node i. before broadcasting ξi (k+1) to neighbors of node i, we obtain
Then, we provide the following situations in which the private the following equation, 
information will be disclosed under CEMA. Pi (k + 1) = ξi (k + 1) − qij ξj (k) + Pi (k) − θi (k)
1) Nodei can observe the neighbor set of its neighbors and j∈V

Nj+ j ⊆ Ni , j = i, i.e., node i can receive all infor- = ξi (k + 1) − qij ξj (k) − θi (k) + ξi (k)
j∈V
mation of neighbors of node j and node j. Such situation 
may happen practically, for example, the communication − qij ξj (k − 1) + Pi (k − 1) − θi (k − 1)
j∈V
topology is complete.   
= ξi (k + 1) − qij ξj (k) − θi (k) + Pi (0)
2) Suppose that there is an eavesdropper, who knows W and
k k j∈V k
Q. Then, the eavesdropper can obtain the private infor- (11)
mation of consumer j by eavesdropping the information Taking limitations on both sides of (11), one obtains
lim Pi (k + 1)
broadcast by node j and neighbors of node j. k→∞
   
The detailed privacy disclosure process is as follows. After = lim [ ξi (k + 1) − qij ξj (k) − θi (k) + Pi (0)]
k→∞
k j∈N + ∪i
initialization and the first time iteration, node i or the eaves- k
i
k
(12)
dropper obtain λj (1) through the communication link, and Since Pi (0) is known and ξj (k), j ∈ Ni+ ∪ i, k = 1, 2, · · ·
Pj (1) according to can be heard through communication links, lim Pi (k) can
 k→∞
ξj (1) = qjl ξl (0) + Pj (1) − Pj (0), (6) be obtained if the sum of the added noises is zero. However,
+ specific Pi (k), ∀k = 1, 2, · · · cannot be obtained, which means
l∈V
as ξl (0) and Pj (0), l ∈ Nj l = j are known. By recursion, that the sensitivity of Pi to λi cannot be inferred. Hence, in
λj (k), Pj (k), k = 1, 2, · · · will be disclosed. Through curve this way, only part of the private information can be preserved.
fitting, the derivative of the utility function of each consumer
and the upper and lower bounds of the power usage can be B. P-CEMA
acquired. Hence, the private information, i.e, the electricity Inspired by [17], we propose the privacy-preserving al-
consumption lim Pj (k) and the sensitivity of the electricity gorithm to preserve the complete privacy of the consumer
k→∞
consumption Pi to the electricity price λi will be disclosed. while keeping the optimality. Differently, we preserve the
Therefore, the interesting problem is how the private informa- final state and the functional relationship between two states
tion of consumers can be preserved while the convergence and of CEMA, which is more complex than average consensus
the optimality of the final solution are guaranteed. considered in [17]. To avoid the disclosure of the private

‹,(((
    
electricity consumption, we introduce a secret continuous Fji (zji ) − Fij (zij ) = 0. (16)
function Fij (z) : R → R, i ∈ Nj+ for node i with respect to i∈V j∈N +
i
i∈V j∈N −
i
its out-neighbor node j. Suppose that Fij (z) is only available From (16), we note that the sum of all noises added to
to node i and j, which can make the initial state Pi (0) ξi (k), ∀i ∈ V, k = 1, · · · is zero and the noise added to ξi (k)
not inferrable specifically. To guarantee the convergence and is exponentially decreasing and goes to zero with k. Hence,
optimality, we design the noise adding process for λi (k) and according to Lemma 1 and 2, the convergence and optimality
ξi (k), ∀i ∈ V , where the absolute bound of the noise is of P-CEMA can be guaranteed.
exponentially decreasing with iterations and the sum of all We then prove that Pi∗ , i ∈ V will not be disclosed
noises at all iterations will converge to zero. The details are specifically. According to P-CEMA, for each node i, there
shown in Algorithm 1. holds, 
lim θi (k) = νi (0) + νi (1) − ν̃i (0) + 2 νi (2) − νi (1) + · · ·
k→∞
Algorithm 1 Privacy-preserving CEMA (P-CEMA) k
 
Initialization: = νi (0) − νi (0) + [ Fij (zij ) − Fji (zji )]
1. Each node i selects a uniform distribution random vari- j∈Ni− j∈Ni+
 
able νi (0) from interval [− √13σ , √13σ ], where σ > 0 is =[ Fij (zij ) − Fji (zji )]
a constant, and arbitrarily gives constant sequences zij j∈Ni− j∈Ni+

(∈ R) for j ∈ Ni− . Meanwhile, node i initializes λi (0), Since Fij (zij ) is only available to node i and j, neighbors
(17)
ξi (0) and Pi (0) according to (3) and sets an appropriate
η and the termination errors
m and
l . node i or eavesdropper
of  cannot know the specific value of
[ j∈N − Fij (zij ) − j∈N + Fji (zji )]. Therefore, neighbors of
2. Each node i sets θi (0) = νi (0) and transmits zij , λi (0) i i
node i or eavesdroppers cannot compute lim Pi (k) exactly
and ξi (0) to its neighboring node j. k→∞
according to (12), implying that the final electricity consump-
3. Each node i calculates ν̃i (0) by,
  tion of each consumer i will not be disclosed.
ν̃i (0) = νi (0) − [ Fij (zij ) − Fji (zji )], ∀i ∈ V. Then, we prove that the sensitivity of the electricity con-
(13)
j∈Ni− j∈Ni+ sumption to the price will not be known explicitly. Due to
Iteration: loop
(11) and Pi (0) = ξi (0) = 0, i ∈ V , we have
4. Each node i generates a uniform distribution random Pi (1) = ξi (1) − qij ξj (0) + Pi (0) − θi (0) = ξi (1) − θi (0) (18)
variables νi (k) from interval [− √13σ , √13σ ] for k ≥ 1. j∈Ni
Then, node i updates

θi (k) according to, As the exact value of θi (0) is unknown, Pi (1) cannot be
νi (1) − ν̃i (0) if k = 1 fixed by obtaining ξj (1), j ∈ Ni+ ∪ i. For the same reason,
θi (k) = (14)
k νi (k) − k−1 νi (k − 1) if k ≥ 2 Pi (k), k = 1, 2, · · · cannot be obtained. Meanwhile, since
where  ∈ (0, 1) is a constant for all nodes. lim Pi (k) is not known, Pi (k), k = 1, 2, · · · cannot be fixed
k→∞
5. Each node i updates λi (k + 1), Pi (k + 1) and ξi (k + 1) backwards. Although we obtain specific λi (k), k = 1, 2, · · · ,
according to (4) and adds the noise θi (k) to ξi (k + 1), the sensitivity of Pi to λi cannot be acquired.
i.e., ξi (k+1) = ξi (k+1)+θi (k). Then, node i broadcasts Remark 1: As we can see that from P-CEMA, the secret
λi (k + 1) and ξi (k + 1) to its neighbors. function is only required at the initialization of the noise
6. If |ξi (k)| ≤
m , ∀i ∈ V and |λi (k)−λi (k −1)| ≤
l , ∀i ∈ θi (0), ∀i ∈ V . Compared with encrypting communication
V , break. variables, utilizing secret function is more simple.
Output: λi (k), Pi (k), ∀i ∈ V .
V. S IMULATIONS
In this section, we illustrate the performance of the proposed
C. Performance Analysis of P-CEMA
privacy-preserving algorithm by extensive simulation results.
We prove that P-CEMA can preserve the privacy of the We consider the IEEE-39 Bus System with ng = 10 gener-
electricity consumption and the sensitivity of the electricity ation units and nd = 18 consumers, where the communication
consumption to the electricity price, while guaranteeing the topology is strongly connected with directed edges. The utility
convergence and optimality. function of each consumer
⎧
j is described by
Theorem 2: Under P-CEMA, there exists a sufficiently ⎨ ωj Pj − αj Pj2 , Pj ≤
ωj
2αj
small η such that lim Pi (k) = Pi∗ , ∀i ∈ V , where Pi∗ , ∀i ∈ V Uj (Pj ) =
⎩ ωj2 ωj . (19)
k→∞ 4αj
, Pj > 2αj
is the optimal solution, while Pi∗ and the sensitivity of Pi to
The parameters of cost functions of generation units and utility
λi , ∀i ∈ V can be preserved.
functions of consumers are referred by [13]. Meanwhile, let
Proof: To prove the optimality, by adding θi (k) for k =
η = 1.786 ∗ 10−3 , σ = 0.1732,  = 0.03. The secret function
1, 2, · · · , i = 1, · · · , n, we will have,
∞ 
     is given as Fij (zij ) = j ∗ zij + i.
θi (k) = Fji (zji ) − Fij (zij ). (15) First, we investigate the convergence and optimality per-
k=1 i∈V i∈V j∈N + i∈V j∈N − formance of P-CEMA compared with that of CEMA. We set
i i

As the communication network is strongly connected, there all zij ’s from the random range [0,5]. It can be observed from
holds Fig. 1 that P-CEMA can achieve almost the same convergence

‹,(((
 3í&(0$ σ 
 2SWLPDOYDOXH
&(0$
quantity of electricity consumption and the random and expo-
3í&(0$ σ  í

Social welfare
 nentially decreasing noise is exploited to preserve the utility
 function inspired by our works in [17]. Illustrating examples
 and extensive simulation results verify the effectiveness of P-

  
k
   CEMA and the theoretical analysis.
Fig. 1. Convergence and optimality of P-CEMA
R EFERENCES
speed compared with CEMA and the optimality can always be
[1] X. Fang, S. Misra, G. Xue and D. Yang, “Smart grid-The new and
guaranteed. By adding noises with a sufficiently large σ, we improved power grid: A survey”, IEEE Communications Surveys and
find that the convergence speed will remain unchanged. When Tutorials, vol. 14, no. 4, 944–980, 2012.
σ is too much small, i.e., the noise will be selected from a [2] P. Siano, “Demand response and smart grids–A survey”, Renewable and
Sustainable Energy Reviews, 461–478, 2014.
large interval, P-CEMA may not reach convergence. [3] N. Li, L. Chen and S. H. Low, “Optimal demand response based on utility
Then, we investigate how the private information can be maximization in power networks”, in Proc. of PESGM, 1–8, 2011.
preserved under P-CEMA with same setting of zij ’s. It can be [4] Z. Fan, “A distributed demand response algorithm and its application to
PHEV charging in smart grids”, IEEE Transactions on Smart Grid, vol. 3,
observed from Fig. 2 that for different kinds of generation units no. 3, 1280–1290, 2012.
and responsive consumers, the final power and the sensitivity [5] A. H. Mohsenian-Rad, V. W. S. Wong, J. Jatskevich, J. Jatskevich, R.
can not be inferred by neighbors or the eavesdropper without Schober and A. Leon-Garcia, “Autonomous demand-side management
based on game-theoretic energy consumption scheduling for the future
knowing the secret function. Hence, the privacy of consumers smart grid”, IEEE Transactions on Smart Grid, vol. 1, no. 3, 320–331,
is preserved under the proposed algorithm while the con- 2010.
vergence and optimality remained unchanged. To investigate [6] Z. Zhang, M. Y. Chow, “Convergence analysis of the incremental cost
consensus algorithm under different communication network topologies
how the different distribution will affect the performance of in a smart grid”, IEEE Transactions on Power Systems, vol. 27, no. 4,
the algorithm, we set zij as zij = 5 ∗ i/j and compare 1761–1768, 2012.
the performance of P-CEMA with the normal and uniform [7] G. Binetti, A. Davoudi, F. L. Lewis, N. David and T. Biagio, “Distributed
consensus-based economic dispatch with transmission losses”, IEEE
distributions and different secret functions. It can be seen in Transactions on Power Systems, vol. 29, no. 4, 1711–1720, 2014.
Fig. 4(a) that different noises will not affect the performance [8] S. Yang, S. Tan and J. X. Xu, “Consensus based approach for economic
of P-CEMA differently, while different secret functions will dispatch problem in a smart grid”, IEEE Transactions on Power Systems,
vol. 28, no. 4, 4416–4426, 2013.
induce more differences shown in Fig. 4(b). [9] H. Xing, Y. Mou, M. Fu and Z. Lin, “Consensus based bisection approach
 
3í&(0$ 3í&(0$ for economic power dispatch”, in Proc. of IEEE CDC, 3789-3794, 2014.
 &(0$
&(0$  [10] S. Kar, G. Hug, J. Mohammadi and M. F. M. Jose, “Distributed State

 Estimation and Energy Management in Smart Grids: A Consensus Inno-
3LN

3 N


vations Approach”, IEEE Journal of Selected Topics in Signal Processing,
L

í
 L  vol. 8, no. 6, 1022–1038, 2014.
í í
L  [11] N. R. Asr, Z. Zhang, M. Y. Chow, “Consensus-based distributed energy
í í
í 
λLN
  í í    management with real-time pricing”, in Proc. of PESGM, 1–5, 2013.
λLN
[12] N. Rahbari-Asr, U. Ojha, Z. Zhang and M. Chow, “Incremental wel-
  fare consensus algorithm for cooperative distributed generation/demand
 L   L 
response in smart grid”, IEEE Transactions on Smart Grid, vol. 5, no. 6,
2836–2845, 2014.
 
[13] C. Zhao, J. He, P. Cheng, and J. Chen, “Consensus-based energy
3LN

3LN

í í
management in smart grid with transmiss ion losses and directed
3í&(0$
í 3í&(0$ í
&(0$
communication”, IEEE Transactions on Smart Grid, DOI: 10.1109/TS-
&(0$
í
í í   
í
í í   
G.2015.2513772, 2016.
λLN λLN [14] A. Mandal, “Privacy Preserving Consensus-Based Economic Dispatch
Fig. 2. Units and consumers in Smart Grid Systems”, in International Conference on Future Network
VI. C ONCLUSION Systems and Security, Springer International Publishing, 98–110, 2016.
[15] H. Li, X. Lin, H. Yang, X. Liang, R. Lu, and X. Shen, “EPPDR: an
In this paper, we mainly consider the privacy-preserving efficient privacy-preserving demand response scheme with adaptive key
problem of the consensus-based algorithm for DEM consider- evolution in smart grid”, IEEE Transactions on Parallel and Distributed
Systems, vol. 25, no. 8, 2053–2064, 2014.
ing multiple generation units and responsive consumers. We [16] S. Han, U. Topcu, G. J. Pappas, “Differentially private distributed
reveal that the private information of consumers including the constrained optimization”, IEEE Transactions on Automatic Control,
electricity consumption and the sensitivity of the electricity DOI: 10.1109/TAC.2016.2541298, 2016.
[17] J. He, L. Cai, C. Zhao, and P. Cheng, “Privacy-preserving Average
consumption to the electricity price can be disclosed from Consensus: Privacy Analysis and Optimal Algorithm Design”, http-
the considered algorithm. To preserve the privacy of the s://arxiv.org/pdf/1609.06368v1.pdf, 2016.
consumers, the secret function is utilized to preserve the [18] J. E. Stiglitz, “Whither socialism?”, MIT press, 1996.
[19] P. Samadi, A. H. Mohsenian-Rad, R. Schober, V. W.S. Wong, and
Fig. 3. Performance comparisons J. Jatskevich, “Optimal real-time pricing algorithm based on utility
  maximization for smart grid”, in Proc. of SmartGridCom, 415–420, 2010.
  [20] C. Zhao, J. He, P. Cheng, and J. Chen, “Analysis of Consensus-
L  L 
  based Distributed Economic Dispatch under Stealthy Attacks”, IEEE
3 N

3 N

  Transactions on Industrial Electronics, DOI: 10.1109/TIE.2016.2638400,

L

í 8QLIRUP í

)LM]LM M ]LM L
2016,
1RQRLVH
í í ) ]  M ] L
1RUPDO LM LM LM
í í
í í í    í í   
λLN λ N
L

(a) Different distribution noises (b) Different secret functions

‹,(((