Quantum computing for energy systems optimization: Challenges

and opportunities

Akshay Ajagekar, Fengqi You *

Cornell University, Ithaca, New York 14853, USA

Abstract

The purpose of this paper is to explore the applications of quantum computing to energy
systems optimization problems and discuss some of the challenges faced by quantum computers
with techniques to overcome them. The basic concepts underlying quantum computation and their
distinctive characteristics in comparison to their classical counterparts are also discussed. Along
with different hardware architecture description of two commercially available quantum systems,
an example making use of open-source software tools is provided as a first step for diving into the
new realm of programming quantum computers for solving systems optimization problems. The
trade-off between qualities of these two quantum architectures is also discussed. Complex nature
of energy systems due to their structure and large number of design and operational constraints
make energy systems optimization a hard problem for most available algorithms. Problems like
facility location allocation for energy systems infrastructure development, unit commitment of
electric power systems operations, and heat exchanger network synthesis which fall under the
category of energy systems optimization are solved using both classical algorithms implemented
on conventional CPU based computer and quantum algorithm realized on quantum computing
hardware. Their designs, implementation and results are stated. Additionally, this paper describes
the limitations of state-of-the-art quantum computers and their great potential to impact the field
of energy systems optimization.

*
Corresponding author. Phone: (607) 255-1162; Fax: (607) 255-9166; E-mail: [email protected]

1
Graphical abstract

Keywords: Quantum computing, energy systems, optimization

List of Abbreviations: CPU, Central Processing Unit; GPU, Graphics Processing Unit; HENS,
Heat Exchanger Network Synthesis; QAOA, Quantum Approximate Optimization Algorithm;
QPU, Quantum Processing Unit; QUBO, Quadratic Unconstrained Binary Optimization; UC, Unit
Commitment; VQE, Variational Quantum Eigensolver

Nomenclature

Ai Zeroth order cost coefficient for unit i

A, B Weight parameters for QUBO
Bi First order cost coefficient for unit i
Ci Second order cost coefficient for unit i
Cij Cost of transporting one unit of energy from location i to location j
cij Cost of heat exchanger between source i and sink j
Dj Heat demand at sink j

2
fi Fuel cost of committed unit i
hi Size of grid for power generated by unit i
L Load requirement
pi Power generated by unit i
Pmin,i Minimum possible power generated by unit i provided unit is online
Pmax,i Maximum possible power generated by unit i provided unit is online
qij Heat transfer between source i and sink j
Si Heat supply for source i
Tpq Number of units of energy transported from plant p to plant q
U Set of units
Uij Maximum possible heat transfer between source i and sink j
vi 0-1 variable representing whether unit i is offline
wij 0-1 variable for heat match between source i and sink j
xpi 0-1 variable for assignment of plant p to location i
yi 0-1 variable representing whether unit i is online
z 0-1 variable denoting grid points
Z Pauli operator assuming values in {-1, 1}

3
1. Introduction
With the rising demand for energy and the need for environmental protection, there has been
a primary interest in the design, control, and planning of energy systems. New energy resources
are also being integrated into energy systems rendering optimal management and control of
available resources to be a key issue in harnessing new technologies. Without optimal utilization
of resources, the cost of investment for these technologies cannot be justified. Therefore,
optimization tools and algorithms provide a suitable way to solve complex energy systems
problems in this field. For the comparison between renewable energy sources, factors such as price
of generated energy, greenhouse gas emissions, availability of resources, efficiency of energy
conversion, natural resource requirement, and social impacts have been previously taken into
account [1]. Proper allocation of available energy sources requires the study of models such as
energy planning models, energy supply-demand models, forecasting models, renewable energy
models, emission reduction models, and optimization models [2]. Energy generation planning and
scheduling systems, location and transportation problems, resource allocation, engineering design,
network planning are some of other areas wherein optimization methods has found widespread
applications [3]. Novel optimization frameworks with comprehensive energy conversion
modelling and complete system optimization is an important tool to reduce life-cycle cost and
evaluate optimal design for micro-grid energy systems [4]. The worldwide energy crisis demands
development of hybrid renewable energy systems instead of fossil fuels. Simulation based studies
of large-scale global optimization for such hybrid energy systems through energy management
strategies have also been conducted [5]. Reliability and cost of power systems with integrated
renewable resources are important aspects considered by system operators. Optimization methods
for maintenance by considering the system reliability can be applied for such systems [6]. Apart
from the deterministic optimization methods, in the last few decades, heuristic approaches and
artificial neural networks have also been proposed. Meta-heuristic algorithms like simulated
annealing [7], tabu search [8], genetic algorithms [9], ant-colony optimization [10] have grown
quite popular due to their ability to tackle complex optimization problems where traditional
methods prove ineffective. As the complexity of some problems increase, exponential computation
time might be needed to find the optimum in the worst case. For example, in the case of power
grid networks, complexity of the network is a function of the number of power plants that are a
part of the grid. According to the data obtained from U.S. energy information administration (EIA),

4
Fig. 1 shows a continuous rise in the number of power plants in the United States for the last
decade, thus complexity of power networks is expected to keep increasing. Hence, there arises a
need for new tools and techniques which are capable of addressing such energy systems
optimization problems with high complexity, and which provide good solutions in reasonable
runtimes.
Quantum computing provides a novel approach to help solve some of the most complex
problems while offering an essential speed advantage over conventional computers [11]. This is
evident from the quantum algorithm proposed by Shor for factorization which is exponentially
faster than any known classical algorithm [12], and Grover’s quantum search algorithm capable of
searching a large database in time square root of its size [13]. Quantum chemistry [14], machine
learning [15], cryptography [16] and optimization are some of the areas where a quantum
advantage is perceived when facilitated by a quantum computer. Recent advances in hardware
technology and quantum algorithms allow for complex energy systems optimization problems to

Fig. 1. Number of power plants in U.S. in the last 11 years

be solved on a quantum computer. Methods such as process integration, superstructure

optimization, and their applications to sustainable design and synthesis of energy systems are
crucial and have been an active research area [17]. These methods due to their high complexity
require classical optimization solvers to solve the formulated problem, which can be
5
computationally expensive with no guarantee of returning a solution. Classical optimization
methods applied to large-scale renewable and sustainable energy systems to perform multi-
objective optimization result in use of high computational effort [18]. Physical modelling and
optimization of hybrid energy power systems require exponential computation time as the size of
problem increases [19]. It is imperative to account for optimal design and multi-scale decisions of
shale gas energy systems that are economical, sustainable and socially responsible. Modelling and
optimization of such shale gas energy systems under multiple types of uncertainties is a
computational challenge [20]. Decomposition of complex large-scale design and synthesis
optimization of energy system problems can result in nested optimization problems simpler than
the original but is much more computationally intensive [21]. An integral part of designing energy
supply chains with multi-scale complexities is to build a large-scale integrated model that includes
minute details across all spatial and temporal scales. Classical general purpose optimization
methods and off-the-shelf optimization solvers fail to compute feasible solutions for the resulting
large-scale multi-objective optimization problem [22]. Optimizing the bioenergy and biofuel
supply chain network structure comprising of multiple sites and multiple echelons is performed at
the design stage through superstructure optimization. However, a comprehensive and detailed
superstructure although appealing, may be computationally intractable [23]. With the increasing
importance of water-energy-food nexus, challenges that account for its multiple scales, appropriate
system boundaries, modelling the decision making and conflicting objectives along with
uncertainties must be considered. Although a higher level of detail provides more thorough details,
the associated modelling and computational challenges increase [24]. Custom algorithms for a
specific problem class can sometimes outperform the best available classical solvers, but a generic
solution approach is much more desirable. While solving large instances of complex problems
with deterministic technique is intractable, approximate algorithms should be considered.
Quantum computers realize such approximate algorithms intrinsically.
Important large-scale problems of practical relevance can consist of thousands of variables,
and constraints which due to their complex combinatorial nature may require days or even weeks
to converge to an optimal solution. Design of shale-gas supply chain network covering more than
10,000 km2 area requires optimizing a mixed-integer linear problem with 51,133 variables and
51,880 constraints. State-of-the-art classical CPU-based solver takes more than 15 hours to
compute a solution with the desired optimality gap [25]. Another such example of an integrated

6
optimization framework for energy supply chain involves optimization of a quadratic constrained
mixed-integer problem with 8,321 total variables and 7,247 constraints. Solving this problem by a
classical CPU-based solver requires 15.6 hours of computation time [26]. Although energy
systems problems have been well acknowledged by industry and academia, most of the existing
studies are limited to regional scales. Extending such studies to national or global scales faces the
limitation of computational power and can result in computationally intractable mathematical
problems. All these challenges motivate the need of developing novel and more efficient solution
strategies for energy systems optimization problems. Quantum computing is a game-changer
technology that is able to tackle such intractable problems and may produce good solutions in
reasonable runtimes. Unlike classical computers exploring the entire space of feasible solutions,
quantum computers focus on exploring selected feasible subspaces, thus inducing a quantum
speedup. This paper exemplifies the promise of quantum computation through well-known energy
systems optimization problems.
The purpose of this paper is to introduce readers to the new and emerging field of quantum
computation for optimization, and state its implications and applications to energy systems
engineering. A brief background unraveling the working of a quantum computer and its major
dissimilarities with conventional CPU/GPU-based classical computers will also be provided.
Conventional or classical computers are governed by classical physics, and simply put are
integrated multi-purpose computers used by most people on a daily basis. Furthermore, examples
solving small optimization problems on different types of quantum computers will be illustrated
here allowing the readers to get a basic idea on using quantum computers for optimization
purposes. Finally, a detailed methodology for application of quantum computers to classes of
energy systems optimization like facility location allocation, unit commitment problems and heat
exchanger network synthesis, is discussed. Some of the challenges faced by quantum computing
for energy systems optimization, as well as its limitations and scope in the future, are also
discussed. The major contributions of this paper are listed as follows:

• A brief review of quantum computing fundamentals and operating functionality of two

types of quantum computing architectures has been provided. A step-by-step tutorial of
developing solution strategies for energy systems optimization problem using both the
above-mentioned quantum computing architectures is put forward.

7
 • Novel reformulations of energy systems optimization problems namely facility location-
allocation, unit commitment and heat exchanger network synthesis into unconstrained
binary optimization problems has been proposed to facilitate ease of mapping and solving
on quantum hardware.
• A performance comparison study in aspects of timing and solution quality is also conducted
for each class of problems using both state-of-the-art quantum computing algorithms and
best available classical CPU-based solution methods.
• While addressing the limitations of proposed methodologies owing to technological
barriers, the future scope of quantum computing for large-scale complex energy systems
optimization problems is discussed and some ideas to harness the full potential of quantum
computers which can pave the way to a new sophisticated solution architecture for energy
systems optimization are also put forth.

2. Quantum Computing Fundamentals

Over the past five decades, Moore’s law has held true with the computing power doubling
every two years by reducing the size of transistors in an integrated circuit. The current state of the
art transistors are only a few atoms thick [27]. However, as this size gets smaller quantum effects
start to interfere with their functionality. Quantum computing is the next frontier for computation.
A broad definition of quantum computing is computing that follows the logic of quantum
mechanics.
Analogous to bit being the fundamental unit of classical computing, quantum computing is
built upon the quantum bit called qubit. A classical bit can be in either of state 0 or 1, while a qubit
is in the superposition state of 0 and 1 together termed as computational basis states. Dirac

notation “ | ” is the standard notation for states in quantum mechanics. Fig. 2 shows the Bloch

sphere representation of a quantum bit which helps to visualize the state of a single qubit. Unlike
classical bit which can be in either of its two states, infinite number of states are possible for a
qubit. After it is measured, the state of qubit collapses to one of the basis states [11]. Another
elegant property of qubits which sets them apart from classical bits is entanglement. Quantum
entanglement is the ability to form co-relations between individually random behaviors of two
qubits. These properties of superposition and ability to exist in entangled states are exploited in a

8
quantum computer resulting in high computation power. A few of proposed quantum computing
architectures have already been realized through technological advancements. A brief overview
on the operation and the applications of these quantum computing technologies is provided below.

Fig. 2. Fundamental units of computing

2.1. Quantum Circuit Model

A quantum circuit comprises of quantum gates manipulating qubits to perform calculations.
Analogous to logic gates in classical digital circuits, quantum gates are the building blocks of
quantum circuits. In this model, quantum gate operations are applied one by one to the state of the
system, thus evolving it towards a desired solution of the problem [28]. Gate model of quantum
computation is an alternate term used for this model. In the quantum circuit, the tasks of preparing
a specific input, applying a set of gate operations and measuring the state of qubits in the
computational basis is carried out sequentially [29]. Quantum algorithms play a vital role in
implementing the power of quantum hardware. Approximate solutions for combinatorial
optimization problems can be produced by Quantum Approximate Optimization Algorithm
(QAOA) [30]. In the field of chemistry and optimization, a classical-quantum hybrid algorithm
called Variational Quantum Eigensolver (VQE) has shown exceptional performance in terms of
speed and resource utilization [31]. Generating a trial state and estimating its energy is performed
on a quantum computer with the energy being optimized on classical computer.
However, practical applications of the gate model quantum computer are limited. Few
problems like integer factorization and solving linear system of equations show exponential

9
speedup [32], while for other problems gate model quantum computers are not known to be faster
than CPU/GPU-based classical computers. The qubits in this model are susceptible to de-
coherence, meaning their quantum states are destroyed by interactions with the environment. A
universal quantum computer can be realized after overcoming these setbacks and joining forces
with classical computing. Tech giants in the field of computation like IBM have already launched
a cloud based commercial version of a gate model quantum computer, and is working towards
building an universal quantum computer [33].

2.2. Quantum Annealing based model

Quantum Annealing, which is essentially the noisy version of quantum adiabatic computation,
is a quantum counterpart of simulated annealing. Instead of thermal fluctuations [7], quantum
fluctuations are introduced in quantum annealing for faster convergence to optimal state [34]. A
quantum annealing machine has wide range of applications, including ability to solve
discrete/combinatorial optimization problems. Unlike the gate circuit model, quantum annealing
machines or annealers are explicitly designed for a specific purpose, like optimization, machine
learning, etc. It has also been proven that an extended form of quantum annealing is theoretically
equivalent to gate (circuit) model [35]. The Canadian company, D-Wave, provides state of the art
quantum annealing machines that have wide range of applications with cloud access to its users
[36].
This model of quantum computer realizes the quantum annealing algorithm by enabling the
quantum tunneling processes. In order to minimize the cost function, a time dependent term
representing quantum fluctuations is added, such that the probability of existence of optimum is
uniform over all possible states. This can be seen in Fig. 3 where the cost function is a double well
with local optimum states and the probability distribution P( x, t ) being uniform at the beginning.
After allowing the natural time evolution of the system, the probability is highest at global optimal
solution at the end of the annealing process. Apart from quantum hardware, this algorithm can also
be simulated on classical computers using a deterministic approach or quantum Monte Carlo
techniques for large problems [37].
Although quantum annealing machines are able to withstand noise from environment, there
are very few problems, which can be solved exponentially faster on these machines as compared
to classical algorithms. Also, in comparison to gate circuit model the number of qubits used in

10
annealers is much higher. Simultaneously, a hardware and language independent approach is being
developed to enable migration of existing scientific code across all architectures [38]. At this stage
of development there is a tradeoff between the above mentioned models of quantum computation
and an informed decision must be made based on the complexity class, size of the problems, and
whether or not a quantum advantage is expected.

Fig. 3. Time evolution of Quantum Annealing process for optimization of a function having both local
and global optimum

3. Facility location-allocation in energy systems optimization

Facility location and allocation problems are essential components of strategic design and
planning for energy systems [39]. Some energy systems infrastructure development optimization
problems could be cast as facility location-allocation problems. There has been substantial research
on their model formulations, and various algorithms have been developed to solve such problems
[40]. Examples of this type of energy systems optimization problems include determining the
optimal location for wind farms to maximize energy capture while accounting for electrical grid
constraints [41]. Locations of facilities such as solar or wind power generation systems or hydro
power plant can also be determined based on minimization of facility opening costs, energy
transportation costs subject to satisfying energy demand subject to resource availability constraints
[39].
One such fundamental problem from the category of facility location is the quadratic
assignment problem [42]. Hub-based network design for electricity storage require a hub facility
location model, which is formulated as a quadratic assignment problem [43, 44]. Some solution
approaches for biofuel supply chain optimization also use quadratic programming models [45].
The quadratic assignment problem being one of the most difficult problems of the NP-hard
complexity class, is widely studied in the combinatorial optimization community. A number of
formulations and solution techniques have been proposed accounting for the continuous interest

11
in this problem [46]. In the context of power grid optimization, n power plants are to be assigned
to n regions in such a way that the cost of interplant transportation is minimized. This is illustrated
for four plants or facilities in Fig. 4. Two n × n matrices C=||Cij||and T=||Tpq|| are given, where Cij
is the cost of transporting one unit of energy from location i to location j and Tpq is the number of
units of energy to be transported from plant p to plant q. Each of the n2 assignment variable is
denoted by binary variable xpi to represent whether plant p is assigned to location i. The assignment
constraints ensure that only one facility is assigned to each location. For simplicity purposes, any
demand constraints and resource availability constraints are not considered here. The Koopmans-
Beckmann formulation [47] of this problem is given in Eq. (1).
n n n n
min ∑∑∑∑ C T
=
q 1=
p 1 =j 1 =i 1
ij pq x pi xqj

n
s.t. ∑x
p =1
pi = 1, ∀i = 1, 2...n (1)

∑x
i =1
pi = 1, ∀p = 1, 2...n

In this section, the ability of different quantum hardware, namely the IBM’s gate model
and D-Wave’s quantum annealers, to solve this combinatorial optimization problem is
demonstrated. Problem instances for quadratic assignment were borrowed from the quadratic
assignment problem library (QAPLIB), which provides a unified testbed for quadratic assignment
problems through a compilation of their best known feasible solutions and best lower bounds [48].
These problem instances are solved on the D-Wave’s 2000Q quantum processor, and its
performance is compared to that of a deterministic solver Gurobi run on a Dell Optiplex system
with Intel(R) Core(TM) i7-6700 3.40GHz CPU and 16 GB of installed memory. The same system
configuration and computing environment was used to solve all of the problem instances
mentioned henceforth for both quantum and classical computers.

12
 Fig. 4. Flow between facilities and their locations

3.1. IBM Q’s quantum computer

IBM Q is an initiative to develop scalable quantum systems with a long term goal to
realize an universal quantum computing system [49]. Their devices can be accessed through an
open-source quantum computing framework called Qiskit [50]. Qiskit has three primary
components, Terra, Aqua, and Aer, and they provide for composing quantum programs at circuit-
level, tools and libraries for building quantum applications, and a high-performance simulator for
quantum circuits, respectively.
The quadratic assignment problem can be converted into a single objective cost function given by
Eq. (2) where the parameter A is a free parameter chosen to be large enough to ensure constraint
satisfaction. In order to map the problem to a quantum computer, an Ising Hamiltonian
corresponding to the objective function must be minimized. The Ising Hamiltonian takes the form
of Eq. (3) where Zi is a Pauli operator assuming the eigenvalues ±1 , analogous to a binary variable
taking values in {−1, 1} . The mapping of objective function to an Ising Hamiltonian is done by
the assignment xij → (1 − Z ij ) / 2 , which yields the final form in Eq. (4).
2
 
2
n n n n
 
=C ( x) ∑∑∑∑ CijTpq x pi xqj + A∑ 1 − ∑ x pi  + A∑ 1 − ∑ x pi  (2)
=
q 1=
p 1 =j 1 =i 1 i  p  p  i 

13
 =Η ∑w Z + ∑w Z Z
i
i i
i< j
ij i j (3)

2
n n
 1 − Z pi   1 − Z qj 
n n   1 − Z pi  
=C ( Z ) ∑∑∑∑ CijTpq    + A∑  ∑   − 1
=
q 1=
p 1 =j 1 =i 1  2  2  i  p  2  
2
(4)
  1 − Z pi  
+ A∑  ∑   − 1
p  i  2  
A code sample written in Python programming language is shown in Fig. 5.a where a
quadratic assignment problem instance is solved using a quantum algorithm. In case of custom
user defined problems, the Ising Hamiltonian needs to be manually generated. Exploring the open-
source codebase for the various Ising translators should serve this purpose. The VQE quantum
algorithm is chosen here among all those available in the Aqua library. A classical optimization
technique and a variational method to calculate approximate wave functions along with an
entangler map which specifies the entanglement of qubits are selected, depending on their specific
properties and relevance to the ultimate goal. Finally, the quantum algorithm can be run on a Qasm
simulator or on IBM Q devices remotely via cloud [51].
Execution of the quantum algorithm produces a probability distribution for possible lowest
energy states or ground states for the Ising Hamiltonian, which correspond to the optimal solutions
of the original cost function. The solution with highest probability tends to be the most likely
global optimum. A binary string corresponding to the solution and its probability is returned after
execution. Fig. 5.b represents a histogram for this probability distribution, which clearly highlights
the most probable solutions for their respective problem instances.

14
 Fig. 5. a) Python code sample showing the use of Qiskit to solve a problem instance, and b) Histogram
representing likelihood of possible solutions

3.2. D-wave’s quantum annealer

D-Wave systems use quantum annealing process to search for solutions to a problem. The D-
Wave 2000Q system contains a quantum processing unit (QPU) with 2048 qubits. To solve hard
problems with quantum computer, D-Wave provides a set of open-source Python tools called
Ocean software development kit [52]. A problem along with the user specified parameters is
submitted to the QPU, and problem solutions corresponding to the optimal configuration of qubits
are returned over the network.
Only problems which can be mapped onto an Ising model in Eq. (4) or a quadratic
unconstrained binary optimization (QUBO) formulation as in Eq. (2) can be solved on a D-Wave
system. The coefficients of all the linear and quadratic terms corresponding to the QUBO form of
a problem instance must be declared beforehand in Python dictionary format [53]. A sample code
using the Ocean software libraries to solve the quadratic assignment problem instance and
likelihood of best feasible solutions is shown in Fig. 6, where the ‘linear’ and ‘quadratic’ terms are
user defined. Dimod is an utility that generates the binary quadratic model using both linear and
quadratic terms provided [54]. Embedding of the problem by mapping it to physical qubits on the
annealer is an important step here. Qbsolv is a decomposing solver which finds the minimum of a

15
large QUBO by partitioning it into smaller sub-problems [55]. This solver can be configured to
solve the QUBO problem instance by a meta-heuristic algorithm, Tabu Search, on a CPU-based
classical computer or by quantum algorithm on D-Wave’s quantum system.

Fig. 6. a) Python code sample showing use of Ocean software libraries to solve a problem instance,
and b) Histogram representing likelihood of possible feasible solutions

Since either of the Ising model or quadratic unconstrained binary optimization model is
compatible with the D-Wave systems, the problem in Eq. (1) needs to be mapped appropriately.
Further simplification of the single objective cost function in Eq. (2) and reducing x2 to x, as x
being a binary variable belonging to {0, 1} yields the final QUBO in Eq. (5) with the value of

( )
parameter A chosen, such that A  max CijTpq , and the constant term 2nA used as an offset. The

QUBO is then mapped onto the Chimera graph architecture of qubits on D-Wave QPU through
the minor-embedding process. The QPU then minimizes the energy of this configuration, thus
locating the lowest energy state of this configuration which corresponds to the minimum of
original cost function. The D-Wave annealers are explicitly built for optimization and sampling
problems, so optimization problems tend to run faster on them than on classical computers.
However, as problem size increases partitioning it into sub-problems might affect the timing
performance and the quality of produced solution.
2nA + ∑∑∑∑ CijTpq x pi xqj + 2 A∑∑∑ x pi xqi + 2 A∑∑∑ x pi x pj − 2 A∑∑ x pi
C ( x) = (5)
i j p q i p q> p p i j >i p i

16
 Table 1. Model statistics of quadratic assignment instance for n facilities and n locations

KB formulation (Eq. 1) QUBO (Eq. 5)

no. discrete variables n2 n2
no. constraints 2n 0

3.3. Results
Problem instances of size ranging from 3 to 20 facilities and locations were solved using both
D-wave’s quantum computer and CPU-based classical computer. Known optimal solutions are
available on the QAPLIB collection [48] for the used problem instances. Solutions given by an
exact deterministic solver Gurobi implemented on Intel(R) Core(TM) i7-6700 3.40GHz CPU were
used for cross validation and for the purposes of benchmarking. Model sizes for quadratic
assignment instances for n locations and n facilities are given in Table 1. This means that the largest
problem with 20 candidate locations and 20 facilities includes 400 binary variables.
Table 2 shows the detailed computational times and objective function values for the quadratic
assignment problem instances. All the instances were solved using both classical Gurobi solver
and quantum Qbsolv solver, and the objective values along with computational times required to
calculate them were noted. Best solutions obtained using the quantum solver are reported in the
computational results. A timeout of 12 hours was set for the Gurobi solver, meaning solver
algorithm would stop the solution process after the specified time limit. Objective function values
at this time limit are reported. Best known solutions for these problem instances are also given in
Table 2 to compare the quality of obtained solutions with both classical and quantum techniques.
As evident from Table 2, the classical solver Gurobi uses up exponentially more time as
problem size increases. However, that is not the case with quantum solver. Smaller sized instances
of the quadratic assignment problem which can be embedded onto the specific Chimera graph of
the quantum processor are solved within less than 0.07 seconds. Time required by these small scale
instances is more or less similar. This is because after the process of embedding the whole QUBO
corresponding to a small problem instance on to the Chimera graph is finished, the computation
time includes only the annealing and post-processing times which are proportional to the rate of
reading samples. On the other hand, when an embedding cannot be found for a larger QUBO
instance, the solver Qbsolv splits the input QUBO instance into sub-QUBOs that fit on the D-
Wave system in use. This classical-quantum hybrid algorithm exploits the complementary
strengths of both tabu search and D-Wave solver [56].

17
 Table 2. Computational results for solving quadratic assignment problem instances with classical and
quantum solvers

Gurobi solver Quantum solver

(single CPU core) (D-wave 2000Q)
No. facilities Best known time(s) obj. fun. time (s) obj. fun.
solution

3 24 1.33 24 0.024 24
4 32 1.48 32 0.062 32
5 58 1.5 58 0.066 58
6 94 1.35 94 0.043 94
8 214 1.96 214 0.127 214
9 264 2.01 264 445.23 264
12 578 325.68 578 1946.12 578
14 1014 42,010.42 1014 1008.7 1026
15 1150 ---* 1160 986.19 1160
17 1732 ---* 1750 921.71 1786
20 2570 ---* 2674 744.76 2640
*
Timeout of 12 hours reached by Gurobi

Fig. 7. Solution quality for quadratic assignment problem instances obtained using quantum solver

18
 Similar to other heuristic techniques, a decline in quality of solution is observed for large
problems, where solution space grows exponentially with the size. The quality of obtained solution
can be represented as deviation from the best available optimal solution. This can be seen in Fig.
7, in which a distribution of percentage deviation from best known optimal solution for each
problem instance is plotted. These results show that good solutions without excessive computation
times can be obtained with quantum solver implemented on D-wave system. In spite of the
percentage deviation being not greater than five percent, a more robust partitioning scheme might
be able to eliminate or lower this deviation.

4. Unit commitment problem

Unit commitment (UC) is one of the most popular and critical optimization problems in the
electrical power industry for the operations of power systems. The UC problem is generally
formulated as a large-scale mixed integer nonlinear problem and solving it is very difficult due to
the nonlinear cost function and the combinatorial nature of set of feasible solutions. It has been
proven that UC is not only NP-hard but also NP-complete, so it is impossible to develop an
algorithm with polynomial computation time to solve it [57]. Various techniques to solve the UC
problem have been proposed based on deterministic approach [58], meta-heuristic methods and
combinatorial approaches [59]. Efficient optimization techniques to solve advanced UC models
when high renewable energy sources are integrated to the power systems, are of critical importance
[60]. With the increasing number of energy sources, UC problem poses a hard challenge making
it crucial to develop effective methodologies to tackle this challenge.
Unit commitment is concerned with minimizing the total operational cost to meet an estimated
electric power demand over a given time horizon, while a number of system and generator
constraints must be met [61]. Fig. 8 clearly illustrates the matching of combination of units to
different loads or power demands in a given time interval. System power balance, spinning
reserves and generation power limits of each unit account for some of the constraints in unit
commitment. From a set of units U the amount of power generated by each unit i to satisfy a given
load requirement L is represented by pi. The fuel cost fi of the committed unit i is usually
formulated as a quadratic polynomial with Ai, Bi, and Ci being the coefficients of this polynomial.
The UC problem for a single time period is mathematically represented by mixed-integer quadratic
programming problem in Eq.(6), where binary variable yi represents whether the corresponding

19
unit i is online. The generated power limits of each unit i are imposed by lower bound Pmin,i and
an upper bound Pmax,i. Here, randomly generated UC problem instances were used for study. After
reformulation, these instances were solved on both D-wave 2000Q quantum processor and on
Intel(R) Core(TM) i7-6700 3.40GHz CPU using Gurobi solver to compare the quality of obtained
solutions.

Fig. 8. Commitment of units to different loads/power demands in each time interval

min ∑f
i∈U
i

s.t. fi = Ai yi + Bi pi + Ci pi 2
∑p
i∈U
i =L (6)

Pmin,i yi ≤ pi ≤ Pmax,i yi ∀i ∈ U
yi ∈ {0,1}

4.1. Implementation
Only an unconstrained quadratic optimization problem can be mapped on to D-wave
system, so the mixed-integer quadratic programming problem in Eq. (6) cannot be directly solved
on quantum computing system. The UC problem must be reformulated into a QUBO model, which
can be achieved by discretizing the problem space. The feasible range of continuous variable pi is
divided into N equally spaced grids between Pmin,i and Pmax,i while ensuring that this variable equals

20
zero if the unit is offline. Since this reformulation is an approximation of the original problem, the
number of grids was chosen such that optimal solution of the reformulated UC is within 0.01% of
exact global optimum obtained using a nonlinear solver. Since the problem in Eq. (6) is convex in
nature, Gurobi solver is used to solve the mixed integer quadratic programming problem. The
reformulated UC given in Eq. (7) is a quadratic binary model and can be easily converted to QUBO
through a single objective cost function corresponding to the discretized problem. Eq. (8) forms
the required QUBO and can hence be embedded on to the D-wave quantum processor. The same
instance is solved multiple times on D-wave QPU to compare the likelihood of obtained solutions
to be a global optimum and choose the best solution.
min ∑f
i∈U
i

P −P 
s.t. hi =  max,i min,i 
 N 
N +1
=pi ∑(P
k =1
min,i + ( k − 1) hi ) zik

fi = Ai (1 − vi ) + Bi pi + Ci pi 2 (7)
N +1
vi + ∑ z=
ik 1 ∀i ∈ U
k =1

∑p
i∈U
i =L

vi , zik ∈ {0,1}

∑∑ ( A + B ( P )
N +1
+ ( k − 1) hi ) + Ci ( Pmin,i + ( k − 1) hi ) zik
2
C (=
y, z ) i i min,i
i∈U k =
1
N +1
+ ∑∑ 2C ( P
i∈U k =
1
i min,i + ( k − 1) hi ) ( Pmin,i + ( m − 1) hi ) zik zim (8)
2 2
 N +1
  N +1

+ A∑  vi + ∑ zik − 1 + B  L − ∑∑ ( Pmin,i + ( k − 1) hi ) zik 
= i∈U  k 1 =  i∈U k 1 
4.2. Results
The model sizes for unit commitment problem with n units and N grids are given in Table 3.
The size of discretized problem increases with the number of grids N. For the largest problem with
12 units, the original nonlinear problem includes 12 binary variables, 12 continuous variables and
25 constraints, while the reformulated problem includes 144 binary variables and only 13
constraints.

21
 The randomly generated problem instances for unit commitment in a single time interval were
solved using both classical and quantum solvers. The computational results for the same are given
in Table 4, where the objective function value and time required are mentioned for each model and
respective solution algorithm used. Best solutions and computational time obtained by quantum
solver are also reported in this table. As mentioned earlier choosing proper value of N ensured a
good optimum solution lying within 0.01% of the global optimum. The number of grids used for
discretization is also noted here for each problem instance. Optimal solution of nonlinear and
discretized UC in Eq. (6) and (7), was calculated by the Gurobi solver using regular CPU on
Intel(R) Core(TM) i7-6700 3.40GHz CPU. For the quantum solver, the QUBO parameter values
A and B in Eq. (8) vary for each instance and are chosen based on the coupling strengths and biases
in the original cost objective function and usually take high values to ensure constraint satisfaction.
Biases and coupling strengths are simply the linear and quadratic terms in the QUBO. Values of A
and B are also reported for each instance. The smallest discretized instance with 27 binary variables
can be embedded on to Chimera architecture of quantum processor with ease. However, as the size
of problem and number of grids increase, the corresponding QUBO cannot be directly embedded
on the D-wave QPU. Larger QUBOs require partitioning into smaller sub-problems or sub-
QUBOs, and can be solved using the Qbsolv utility. The partitioning scheme reduces the likelihood
of returning global optimal solutions and might return good enough solutions. Hence, for large
instances, solutions obtained through quantum algorithm exhibit deviation from solutions obtained
using the global optimization solver Gurobi.

Table 3. Model statistics for unit commitment problem with n units and N grids

Non-linear Discretized model QUBO

model Eq. (7) Eq. (8)
Eq. (6)
Number of discrete variables n n(N+2) n(N+2)
Number of continuous variables n 0 0
Number of constraints 2n+1 n+1 0

22
 Table 4. Computational results for solving unit commitment problem using both classical and quantum
solvers

Gurobi solver Quantum solver

(single CPU core) (D-wave 2000Q)
Non-linear model Discretized model
no. obj. time (s) Grids % Error obj. time (s) A B obj. time (s)
units fun. fun. fun.
3 25.5 0.024 7 0.00% 25.5 0.016 2000 5 25.5 0.03
5 13.9 0.028 10 0.00% 13.9 0.02 1000 10 13.9 0.07
8 22.5 0.026 10 0.00% 22.5 0.022 3000 5 25.8 385.75
12 44.49 0.036 10 0.01% 44.5 0.03 6000 5 53.35 1138.16

For a large discretized UC problem solved with Gurobi, increasing the number of grids reduces
the deviation from global optimal solution. Ideally, a similar behavior should be observed for the
case of quantum solver. But high number of grids require higher precision on the quantum system.
Due to the precision limitations for biases and coupling strengths on the D-wave quantum system,
discretizing problem space into smaller grid sizes does not guarantee a better solution. Dividing
the space into grids beyond a certain limit would be redundant and solution quality would further
deteriorate. Thus, there is a trade-off between precision and likelihood of global optimal solution
when solving quantum optimization problems.

5. Heat exchanger network synthesis

In order to avoid excessive energy consumption in industrial plants, energetic integration of
processes is necessary. Heat exchanger network synthesis (HENS) is a systematic tool to control
the costs of energy for a process. Heating and cooling costs are controlled by hot and cold utilities
as well as the heat exchangers in a network. Enhancing heat transfer in heat exchanger devices
using various techniques can also contribute to reduction of these costs and ensure efficient energy
conversion [62, 63]. A cold stream in a process can be either heated by hot utility or hot stream in
the process while a hot stream can be cooled by cold utility or another cold stream. Heat exchanger
network in Fig. 9 comprising of 2 hot streams and 3 cold streams integrated with 4 heat exchangers
illustrates this. Owing to the need for improvement in energy recovery and minimizing the global
cost, heat exchanger network synthesis problem has been extensively studied in the past [64].
HENS is a NP-hard problem for which an exact polynomial-time computational algorithm does

23
not exist [65]. Several metaheuristic techniques capable of handling complex problems in a
feasible computational time have also been proposed [66, 67].
Sequential synthesis and simultaneous synthesis are the two well-known groups of
methods for handling HENS problem. One of the prevalent approaches is sequential synthesis,
wherein a problem is decomposed into three subproblems, namely the minimum utility cost,
minimum number of matches, and minimum cost network problems. These subproblems can be
treated in an easier fashion as compared to the original single-task problem. The minimum number
of matches problem consists of determining minimum cost of matches, which satisfy the supply
and demand of heat for a heat exchanger network. This simple HENS subproblem with only one
temperature interval is NP-hard in the strong sense [65]. For a heat exchanger network consisting
of m sources and n sinks with a single temperature interval, there is a supply Si of heat at each
source i and demand Dj for heat at each sink j. The cost for each possible match between a source
i and a sink j is denoted by cij. The variables qij represent total heat exchanged between source i
and sink j, while the existence of heat match between them is given by a binary variable wij. Fig. 9
clearly shows the existence of heat exchanger match and heat flow between hot stream i and cold
stream j. This matches problem can be formulated as in Eq. (9) where the first two constraints
represent energy balance, and the third one is a logical constraint. Here a randomly generated
problem instance of size m=4 and n=3 was used for analysis that satisfied the feasibility
requirement ∑𝑖𝑖 𝑆𝑆𝑖𝑖 = ∑𝑗𝑗 𝐷𝐷𝑗𝑗 . A pictorial depiction of the same is shown in Fig. 10a. This problem is
reformulated and solved on D-wave’s 2000Q quantum processor and on Intel(R) Core(TM) i7-
6700 3.40GHz CPU using Gurobi solver for solution quality comparison.
min ∑∑ c w
i j
ij ij

s.t. ∑ q=
j
ij Si ∀=
i 1, 2...m

∑ q=
i
ij Dj ∀
=j 1, 2...n

U ij = min{Si , D j } (9)
qij ≤ U ij wij ∀i, j
qij ≥ 0 ∀i, j
wij ∈ {0,1} ∀i, j

24
 Fig. 9. Heat exchanger network containing 2 hot streams and 3 cold streams

5.1. Implementation
The minimum number of matches problem cannot be directly mapped on a quantum computer.
As quantum computers support only discrete problems, the continuous variables qij are discretized
into N equally-spaced grid points in the interval [0, Uij]. The upper bound on qij is set such that the
logical constraint is not violated. Both the original problem and this discretized problem given by
Eq. (10) is solved using the same Gurobi solver to check the viability of the reformulation. The
discrete optimization problem can now be mapped on to D-wave QPU by converting it into a
quadratic unconstrained binary optimization problem. For QUBO a single objective cost function
is given by Eq. (11) containing only binary variables z. Here the value of N was chosen such that
both the mixed-integer linear problem (MILP) and discretized formulations in Eq. (9) and Eq. (10)
respectively produce the optimal objective values with least deviation between them. QUBO
corresponding to the same discretized problem instance is solved on D-wave QPU repeatedly and
compared to the exact solution produced by the classical deterministic optimization solver Gurobi
on a CPU-based computer.

25
 min ∑∑ c w
i j
ij ij

s.t. ∑ q=
j
ij Si ∀
=i 1, 2...m

∑ q= i
ij Dj ∀
=j 1, 2...n

U ij = min{Si , D j } (10)
U ij N
=qij
N
∑ kz
k =1
ijk ∀i, j
N

∑ zijk wij
=
k =1
∀i, j

zijk , wij ∈ {0,1} ∀i, j , k

2
 
2
 N
 N U
z ) ∑∑ cij wij + A  wij − ∑ zijk  + B ∑  Si − ∑∑
C (=
ij
kzijk 
=
i j  k 1=  i  j k 1 N 
2
(11)
 U N

+ B ∑  D j − ∑∑ ij kzijk 
j  i k =1 N 
5.2. Results
Table 5 shows the statistics for model sizes of given formulations of minimum number of
matches problem. For the randomly generated problem instances, MILP model and the discretized
model are solved using CPU-based Gurobi solver. Its corresponding QUBO is solved on quantum
solver and the computation times along with obtained solution are reported in Table 6. For the
largest instance with 15 sources and 15 sinks, the original MILP problem consists of 225
continuous and binary variables while the discretized reformulation has 1350 binary variables. The
number of grid points N was chosen such that the discretization scheme produced an optimal
solution close to the one obtained by solving the original minimum number of matches problem
using an exact solver Gurobi. Choosing the value of N also depends on the viability of solving
very large QUBOs on quantum hardware. Fixing the number of grid points N ensures the existence
of optimal solution close to global minima in the discretized problem space. Since the large
instances cannot be directly embedded on to the Chimera graph architecture of quantum processor,
the decomposing solver Qbsolv is used to partition the large QUBO into smaller sub-QUBOs.
Values of parameters A and B of QUBO in Eq. (11) are set by trial-and-error, while satisfying
A, B  max(cij ) . Here A=20 and B=5 are used.

26
 As evident from the computational results, smaller problem instances solved on quantum
hardware return global optimal solutions. The best obtained solution shown in Fig. 10b exactly
matches the one produced by Gurobi for discretized UC on Intel(R) Core(TM) i7-6700 3.40GHz
CPU. For the largest problem instance, larger values of penalty weight B in the QUBO tend to
violate the energy balance constraints. Larger HENS problem instances grow quickly with the
number of grids and tend to perform poorly given the limited scalability of available quantum
hardware. This is due to the fact that a finite precision is available on quantum annealers, and large
B values result in large coupling strengths and biases. Scaling down these strengths becomes
necessary resulting in loss of precision, thus leading to constraint violation. A simple workaround
for this issue would be to use appropriate number of grids. Also, sharing the edge weight of
Chimera graph can enhance the precision at the expense of extra qubits. A small scale MILP model
translates to a large discretized model, so Gurobi solver performs better than quantum solver with
respect to time and solution quality. Partitioning of the problem into sub-QUBOs also reduces the
likelihood of obtaining global optimal solutions. Improvement in solving large-scale problems on
quantum processor is highly dependent on the scalability and precision offered by quantum
systems. Alternative embedding schemes for QUBO on to quantum processor may facilitate some
form of error mitigation thus improving solution quality. MILPs on CPU-based Gurobi solver have
been optimized to run through parallel computation and perform exceptionally on conventional
computers. Quantum computers on the other hand due to their small scalability and limited
precision do not perform as good as classical MILP solvers. In the context of MILP solution
approaches, quantum computers will require much more technological maturity to compete against
their highly evolved classical counterparts.

Table 5. Model statistics for minimum matches problem with m sources, n sinks and N grid points

MILP Discretized model QUBO

model Eq. (10) Eq. (11)
Eq. (9)
Number of discrete variables mn mn(N+1) mn(N+1)
Number of continuous variables mn 0 0
Number of constraints m+n+2mn m+n+mn 0

27
 Table 6. Computational results for solving heat exchanger network synthesis problems using both
classical and quantum solvers

Gurobi solver Quantum solver

(single CPU core) (D-wave 2000Q)
MILP model Discretized model
no. no. obj. fun. time (s) Grids Error* obj. fun. time (s) obj. fun. time (s)
sources sinks

2 2 5 0.005 4 0.00% 5 0.009 5 0.03

3 3 9 0.015 4 0.00% 9 0.009 9 0.08
4 3 9 0.009 5 0.00% 9 0.014 9 261.6
10 9 20 0.025 5 0.00% 20 0.058 21 519.2
15 15 26 0.049 5 0.00% 26 0.106 26 1288.3
*Relative error between solutions obtained by discretized model and the global optimum

Fig. 10. a) Minimum number of matches problem and b) Obtained optimal solution for the problem
instance

6. Discussion
D-wave quantum devices exhibit limitations in terms of problem size, connectivity of qubits
and precision. The full speed of the machine can only be implemented, if the decision variables in
a problem can be mapped to nodes on Chimera graph architecture of QPU. This process called
minor embedding limits the scope of D-wave quantum machine. Minor embedding in itself maybe
a NP-complete problem and would require an exponential preparation steps even if the hardware
solved a problem in polynomial time. The limited connectivity offered by the QPU demands

28
decomposing a large problem into smaller ones without a guarantee of best solutions. Unlike
classical CPU/GPU-based computers, D-wave devices have very limited precision and this could
cause scaling issues. Presence of intrinsic control errors in these systems and their lack of error
correction negatively impacts the performance of quantum annealers. While quantum annealing
based devices have shown performance gains for specific applications, algorithms for classical
computers can be usually optimized for a specific problem enabling them to outperform quantum
annealers at the current stage. This is evident from the large quadratic assignment instance
performing better than a simpler heat exchanger network synthesis problem in terms of speed and
solution quality. Even in case of a problem for which quantum speedup is apparent, there is no
denying the possibility that a better classical algorithm will be found for the same class of
problems.
The gate based quantum systems achieved by firms like IBM have very few qubits compared
to the annealing systems. Different error sources add noise to qubits in turn affecting the robustness
of gate operation. Current state of the art gate based quantum computer contains 20 qubits with
gate error rate of about five percent. It is projected that for a quantum computer to achieve some
task that no classical computer can perform will need over fifty qubits with less than 0.1 percent
gate error rate. Improving the quality of qubits and qubit operations, as the number of qubits and
associated controls rise, is a major challenge. With increasing number of qubits and their variance,
an important goal would be to mitigate noise from different sources. It is expected that this
challenge would likely get harder as the system size increases. At the current state of technological
advancement, classical algorithms dominate the quantum algorithms implemented on gate systems
with very few exceptions. On the other hand, quantum annealers developed explicitly for
optimization perform better in some cases. This does not necessarily imply that simpler problems
easily solvable on classical computer will perform exceptionally better on quantum hardware.
Problems relevant to practical applications do not always belong to the class of binary quadratic
model, rather a huge chunk of them are mixed-integer nonlinear optimization problems. Until a
universal quantum computer is realized, it is only logical to use classical and quantum hardware
in conjunction with each other. The deterministic methodology of exact classical solvers when
combined with the probabilistic nature of quantum techniques may be the next step in quantum
computing. A crude example of this would be to use quantum resources to explore neighboring
solutions periodically dictated by a classical algorithm to optimize an arbitrary nonlinear objective

29
function. Best of both classical and quantum domains is ensured through this approach. In the near
future, along with better quality and error correction schemes for qubits and controls, there arises
a need for hybrid tools capable of combining the computing power of both classical and quantum
computers.
As evident from the presented case studies, quantum computers perform exceptionally better
for facility-location location allocation problems. Although they perform mediocre for problems
requiring discretization schemes, good feasible solutions can be obtained for problems with large
solution spaces in reasonable runtimes. Sustainable design and operation of energy systems which
consider economic, environmental, and social impacts are quite complex to solve [21], and a
standalone quantum solution approach might not be feasible for such design problems, since the
best solution is always expected in this case. However, for energy systems in the presence of
uncertainty, online adaptation is required to ensure system is always operated optimally, implying
the online solving of such optimization problems. Quantum computers can efficiently handle such
problems, by providing good feasible solutions within reasonably shorter computational times than
any classical computer. However, this does not necessarily mean that design problems will not
benefit from quantum approaches. Hybrid quantum-classical methodologies can be developed for
optimal design and scheduling of energy systems, which will always outperform a either of
classical or standalone quantum approaches. At this stage, such algorithms need to be developed
for a specific class of problems, but a more generic approach for global optimization of highly
complex problems has become a necessity, and efforts need to be directed toward such
technologies and algorithms.

7. Conclusion
In this paper, a brief introduction to the new and emerging field of quantum computing and
some relevant applications to energy systems optimization are provided. The type of problems
compatible with quantum hardware and quantum algorithms used for optimization are also
discussed. The conducted analysis of mentioned optimization problems reveal that although an
approximation by problem space discretization yields an optimal solution, its likelihood of being
the best possible solution changes. Some of the challenges faced by quantum computers in terms
of hardware architecture, precision and error mitigation are also stated. Although in some instances
a quantum advantage may be perceived, classical algorithms explicitly customized for that
particular instance could potentially outperform a quantum computer. This does not imply that

30
quantum computers are up to no good. Quantum computing is at its earliest stages of development
and still has a long way to go compared to its much matured classical counterpart. In the future,
numerous applications would demand implementation of both classical and quantum resources
and hence efforts need to be directed towards harnessing the power of quantum computing systems
for large-scale, complex energy systems optimization problems. Problems like energy supply
planning and supply chain design [22], multi-period scheduling of operations on power plants, and
sustainable design and synthesis of energy systems [17] which can be further complicated by
uncertainties that carry temporal correlations are examples of large-scale optimization problems
for which the said quantum-classical hybrid methodologies should be implemented.

Acknowledgements
This work was supported in part by Cornell University’s David R. Atkinson Center for a
Sustainable Future.

List of References
[1] A. Evans, V. Strezov, and T. J. Evans, "Assessment of sustainability indicators for
renewable energy technologies," Renewable and Sustainable Energy Reviews, vol. 13, pp.
1082-1088, 2009.
[2] S. Jebaraj and S. Iniyan, "A review of energy models," Renewable and Sustainable Energy
Reviews, vol. 10, pp. 281-311, 2006.
[3] A. A. Bazmi and G. Zahedi, "Sustainable energy systems: Role of optimization modeling
techniques in power generation and supply - A review," Renewable and Sustainable
Energy Reviews, vol. 15, pp. 3480-3500, 2011.
[4] T. Cao, Y. H. Hwang, and R. Radermacher, "Development of an optimization based design
framework for microgrid energy systems," Energy, vol. 140, pp. 340-351, Dec 1 2017.
[5] R. L. Tang, X. Li, and J. G. Lai, "A novel optimal energy-management strategy for a
maritime hybrid energy system based on large-scale global optimization," Applied Energy,
vol. 228, pp. 254-264, Oct 15 2018.
[6] E. Shayesteh, J. Yu, and P. Hilber, "Maintenance optimization of power systems with
renewable energy sources integrated," Energy, vol. 149, pp. 577-586, Apr 15 2018.
[7] A. G. Nikolaev and S. H. Jacobson, "Simulated Annealing," in Handbook of
Metaheuristics, 2010, pp. 1-39.
[8] M. Gendreau and J.-Y. Potvin, "Tabu Search," in Handbook of Metaheuristics, 2010, pp.
41-59.
[9] C. R. Reeves, "Genetic Algorithms," in Handbook of Metaheuristics, 2010, pp. 109-139.
[10] M. Dorigo and T. Stützle, "Ant Colony Optimization: Overview and Recent Advances," in
Handbook of Metaheuristics, 2010, pp. 227-263.

31
[11] M. A. Nielsen and I. L. Chuang, Quantum Computation and Quantum Information.
Cambridge University Press, 2010, p. 702.
[12] P. W. Shor, "Algorithms for quantum computation: discrete logarithms and factoring,"
presented at the Proceedings 35th Annual Symposium on Foundations of Computer
Science, 1994.
[13] L. K. Grover, "Quantum Mechanics Helps in Searching for a Needle in a Haystack,"
Physical Review Letters, vol. 79, pp. 325-328, 1997.
[14] B. P. Lanyon et al., "Towards quantum chemistry on a quantum computer," Nature
Chemistry, vol. 2, pp. 106-111, 2010.
[15] J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, "Quantum
machine learning," Nature, vol. 549, pp. 195-202, 2017.
[16] V. Mavroeidis, K. Vishi, M. D. Zych, and A. Jøsang, "The Impact of Quantum Computing
on Present Cryptography," International Journal of Advanced Computer Science and
Applications, vol. 9, 2018.
[17] J. Gong and F. Q. You, "Sustainable design and synthesis of energy systems," Current
Opinion in Chemical Engineering, vol. 10, pp. 77-86, Nov 2015.
[18] R. Banos, F. Manzano-Agugliaro, F. G. Montoya, C. Gil, A. Alcayde, and J. Gomez,
"Optimization methods applied to renewable and sustainable energy: A review,"
Renewable & Sustainable Energy Reviews, vol. 15, no. 4, pp. 1753-1766, May 2011.
[19] B. Bhandari, K.-T. Lee, G.-Y. Lee, Y.-M. Cho, and S.-H. Ahn, "Optimization of hybrid
renewable energy power systems: A review," International Journal of Precision
Engineering and Manufacturing-Green Technology, vol. 2, no. 1, pp. 99-112, January 01
2015.
[20] J. Y. Gao and F. Q. You, "Design and optimization of shale gas energy systems: Overview,
research challenges, and future directions," Computers & Chemical Engineering, vol. 106,
pp. 699-718, Nov 2 2017.
[21] C. A. Frangopoulos, M. R. Von Spakovsky, and E. Sciubba, "A brief review of methods
for the design and synthesis optimization of energy systems," International Journal of
Applied Thermodynamics, vol. 5, pp. 151-160, 2002.
[22] D. J. Garcia and F. You, "Supply chain design and optimization: Challenges and
opportunities," Computers & Chemical Engineering, vol. 81, pp. 153-170, 2015/10/04/
2015.
[23] D. J. Yue, F. Q. You, and S. W. Snyder, "Biomass-to-bioenergy and biofuel supply chain
optimization: Overview, key issues and challenges," Computers & Chemical Engineering,
vol. 66, pp. 36-56, Jul 4 2014.
[24] D. J. Garcia and F. Q. You, "The water-energy-food nexus and process systems
engineering: A new focus," Computers & Chemical Engineering, vol. 91, pp. 49-67, Aug
4 2016.
[25] D. C. Cafaro and I. E. Grossmann, "Strategic planning, design, and development of the
shale gas supply chain network," AIChE Journal, vol. 60, no. 6, pp. 2122-2142, 2014.
[26] O. J. Guerra, A. J. Calderon, L. G. Papageorgiou, J. J. Siirola, and G. V. Reklaitis, "An
optimization framework for the integration of water management and shale gas supply
chain design," Computers & Chemical Engineering, vol. 92, pp. 230-255, Sep 2 2016.
[27] M. Cetina et al., "MoS2 transistors with 1-nanometer gate lengths," Science, vol. 354, pp.
96-99, 2016.

32
[28] L. Gyongyosi, "Quantum Circuit Designs for Gate-Model Quantum Computer
Architectures," 2018, Available: arxiv.org/pdf/1803.02460.pdf.
[29] D. Bacon, "The Quantum Circuit Model and Universal Quantum Computation," in
"Lecture Notes, University of Washington," 2006.
[30] E. Farhi, J. Goldstone, and S. Gutmann, "A Quantum Approximate Optimization
Algorithm," 0471419028, 2014, Available: arxiv.org/abs/1411.4028.
[31] A. Peruzzo et al., "A variational eigenvalue solver on a photonic quantum processor,"
Nature Communications, Article vol. 5, p. 4213, 07/23/online 2014.
[32] A. W. Harrow, A. Hassidim, and S. Lloyd, "Quantum algorithm for linear systems of
equations," Physical Review Letters, vol. 103, 2009.
[33] D. Castelvecchi, "IBM's quantum cloud computer goes commercial," Nature, vol. 543, pp.
159-159, 2017.
[34] T. Kadowaki and H. Nishimori, "Quantum annealing in the transverse Ising model,"
Physical Review E - Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary
Topics, vol. 58, pp. 5355-5363, 1998.
[35] D. Aharonov, W. van Dam, J. Kempe, Z. Landau, S. Lloyd, and O. Regev, "Adiabatic
Quantum Computation is Equivalent to Standard Quantum Computation," SIAM Journal
on Computing, vol. 37, pp. 166-194, 2007.
[36] D.-W. Systems, "Quantum Computing for the Real World Today," 2000.
[37] D. Ceperley and B. Alder, "Quantum monte carlo," Science, vol. 231, pp. 555-60, 1986.
[38] A. J. McCaskey, E. F. Dumitrescu, D. Liakh, M. Chen, W. Feng, and T. S. Humble, "A
language and hardware independent approach to quantum–classical computing,"
SoftwareX, vol. 7, pp. 245-254, 2018.
[39] K. Gokbayrak and A. Selin Kocaman, "A Distance-Limited Continuous Location-
Allocation Problem for Spatial Planning of Decentralized Systems," Computers &
Operations Research, vol. 88, pp. 15-29, 2016.
[40] L. V. Snyder, "Facility location under uncertainty: a review," IIE Transactions, vol. 38, pp.
547-564, 2006.
[41] H. Cetinay, F. A. Kuipers, and A. N. Guven, "Optimal siting and sizing of wind farms,"
Renewable Energy, vol. 101, pp. 51-58, 2017.
[42] R. E. Burkard, E. Çela, P. M. Pardalos, and L. S. Pitsoulis, "The Quadratic Assignment
Problem," in Handbook of Combinatorial Optimization, 1998, pp. 1713-1809.
[43] M. E. O'Kelly, "Hub facility location with fixed costs," Papers in Regional Science, vol.
71, pp. 293-306, 1992.
[44] D. Boutacoff, "Energy Storage Emerging Strategies for Energy Storage," IEEE Power
Engineering Review, vol. 9, pp. 3-6, 1989.
[45] Y. Bai, Y. Ouyang, and J.-S. Pang, "Biofuel supply chain design under competitive
agricultural land use and feedstock market equilibrium," Energy Economics, vol. 34, pp.
1623-1633, 2012.
[46] E. M. Loiola, N. M. M. de Abreu, P. O. Boaventura-Netto, P. Hahn, and T. Querido, "A
survey for the quadratic assignment problem," European Journal of Operational Research,
vol. 176, pp. 657-690, 2007.
[47] G. Abd, E.-N. A. Said, A. M. Mahmoud, and E.-S. M. El-Horbaty, "A Comparative Study
of Meta-heuristic Algorithms for Solving Quadratic Assignment Problem," International
Journal of Advanced Computer Science and Applications, vol. 5, 2014.

33
[48] R. E. Burkard, E. Cela, and S. E. Karisch. (1997). QAPLIB Problem Instances and
Solutions. Available: http://anjos.mgi.polymtl.ca/qaplib/inst.html
[49] (2019, March 30). Quantum computing at IBM. Available:
https://www.research.ibm.com/ibm-q/learn/what-is-ibm-q/
[50] (2019). Qiskit - Quantum Information Science Kit. Available: https://qiskit.org/
[51] "IBM Q Experience Documentation," 2019, Available:
https://quantumexperience.ng.bluemix.net/qx/experience.
[52] (2019). D-Wave's Ocean Software. Available: https://ocean.dwavesys.com/
[53] (2018). D-Wave System Documentation. Available:
https://docs.dwavesys.com/docs/latest/index.html
[54] (2019). Dimod - Dimod 0.8.2 Documentation. Available:
https://docs.ocean.dwavesys.com/projects/dimod/en/latest/
[55] "qbsolv - qbsolv documentation," 2019.
[56] M. Booth, S. P. Reinhardt, and A. Roy, "Partitioning Optimization Problems for Hybrid
Classical/Quantum Execution," 2017.
[57] X. Guan, Q. Zhai, and A. Papalexopoulos, "Optimization based methods for unit
commitment: Lagrangian relaxation versus general mixed integer programming,"
presented at the 2003 IEEE Power Engineering Society General Meeting (IEEE Cat.
No.03CH37491), 2003.
[58] H. Zheng, J. Jian, L. Yang, and R. Quan, "A deterministic method for the unit commitment
problem in power systems," Computers & Operations Research, vol. 66, pp. 241-247,
2016.
[59] A. Bhardwaj, V. K. Kamboj, V. K. Shukla, B. Singh, and P. Khurana, "Unit commitment
in electrical power system-a literature review," presented at the IEEE International Power
Engineering and Optimization Conference, 2012.
[60] S. Y. Abujarad, M. W. Mustafa, and J. J. Jamian, "Recent approaches of unit commitment
in the presence of intermittent renewable energy resources: A review," Renewable and
Sustainable Energy Reviews, vol. 70, pp. 215-223, 2017.
[61] N. P. Padhy, "Unit Commitment - A Bibliographical Survey," IEEE Transactions on
Power Systems, vol. 19, pp. 1196-1205, 2004.
[62] J. Zhang, X. Zhu, M. E. Mondejar, and F. Haglind, "A review of heat transfer enhancement
techniques in plate heat exchangers," Renewable and Sustainable Energy Reviews, vol.
101, pp. 305-328, 2019.
[63] T. Alam and M.-H. Kim, "A comprehensive review on single phase heat transfer
enhancement techniques in heat exchanger applications," Renewable and Sustainable
Energy Reviews, vol. 81, pp. 813-839, 2018.
[64] T. Gundepsen and L. Naess, "The synthesis of cost optimal heat exchanger networks: An
industrial review of the state of the art," Computers & Chemical Engineering, vol. 12, pp.
503-530, 1988.
[65] K. C. Furman and N. V. Sahinidis, "Computational complexity of heat exchanger network
synthesis," Computers and Chemical Engineering, vol. 25, pp. 1371-1390, 2001.
[66] J. M. Ponce-Ortega and L. G. Hernández-Pérez, "Metaheuristic Optimization Programs,"
in Optimization of Process Flowsheets through Metaheuristic Techniques, 2019, pp. 27-
51.
[67] D. Toimil and A. Gómez, "Review of metaheuristics applied to heat exchanger network
design," International Transactions in Operational Research, vol. 24, pp. 7-26, 2017.

34