Académique Documents
Professionnel Documents
Culture Documents
5, SEPTEMBER 2017
Abstract—Advanced metering infrastructure (AMI) is an Security issues in networks can be summarized in terms
important component for a smart grid system to measure, collect, of three main objectives: confidentiality, integrity and avail-
store, analyze, and operate users consumption data. The need of ability [2]. Among various types of threats in the smart grid,
communication and data transmission between consumers (smart
meters) and utilities make AMI vulnerable to various attacks. In distributed denial-of-service (DDoS) is a typical attack that
this paper, we focus on distributed denial of service attack in the severely threatens availability of the communication network
AMI network. We introduce honeypots into the AMI network resources. DDoS attack refers to any event that can reduce or
as a decoy system to detect and gather attack information. We eliminate the proper execution of the network [3], by making
analyze the interactions between the attackers and the defenders, the resources inaccessible for legitimate users.
and derive optimal strategies for both sides. We further prove the
existence of several Bayesian-Nash equilibriums in the honeypot Honeypot based approach is one of the attractive alterna-
game. Finally, we evaluate our proposals on an AMI testbed in tives to counter DDoS attacks as it can protect the network
the smart grid, and the results show that our proposed strat- while also consuming less resources. Honeypots are security
egy is effective in improving the efficiency of defense with the resources that help attract, detect, and gather attack infor-
deployment of honeypots. mation. By pretending to be normal servers to attract the
Index Terms—Honeypot, game theory, advanced metering attackers, honeypots can consume attackers’ resources and
infrastructure, distributed denial of service attack, smart grid. time. They can also influence and interfere with the choice
of intruders, and further detect the intruders’ attack intention.
Other than production systems, the main system can monitor
I. I NTRODUCTION any suspicious intrusion to honeypots.
However, the existing work using honeypots mainly
DVANCED Metering Infrastructure (AMI) is an
A integration of many technologies that provide apt
interactions between client terminals and third party systems.
addresses static defense, which is insufficient to deal with
dynamic attacks [4]. Dynamic attack is a persistent attempt
to introduce invalid data into a system, and/or to damage or
AMI is a crucial component for consumers to obtain near-real
destroy data already stored in it. In addition, as a rational
time price information, which helps them optimize their power
attacker, it is typical that the attackers generally understand
usage. Moreover, AMI makes it possible for the grid to timely
the defense systems in the network by sniffing beforehand.
receive valuable information about consumers [1], e.g., their
Anti-honeypot is used by attackers to identify and detect the
power consumption, aiming at ensuring and enhancing the
defense systems. The attacker can first utilize an anti-honeypot
reliability of the power system. Nonetheless, the two-way com-
to detect the honeypot proxy server in the target network by
munications between the grid and the users may also increase
transmitting initiative packets. Once the honeypot server deter-
the vulnerability of an AMI network to malicious attacks.
mines, the attacker can bypass the honeypot, and access to
the target network through other channels. If the attackers use
Manuscript received August 31, 2016; revised November 22, 2016 anti-honeypots to detect the defense systems in the network
and January 7, 2017; accepted February 5, 2017. Date of publication successfully, they can still find the optimal attack strategies.
February 16, 2017; date of current version August 21, 2017. This work was
supported in part by the NSFC under Grant 61572262, Grant 61533010, Grant In this paper, we study DDoS attacks in AMI networks and
61373135, Grant 61571233, and Grant 61532013, in part by the National introduce a Bayesian honeypot game model. We derive and
China 973 Project under Grant 2015CB352401, in part by the NSF of Jiangsu prove that the equilibrium conditions can be achieved between
Province under Grant BK20141427, in part by the Open Research Fund of the
Key Laboratory of Broadband Wireless Communication and Sensor Network legitimate users and attackers, for the strategies of honeypots
Technology (NUPT), the Ministry of Education under Grant NYKL201507, and anti-honeypots, respectively. As a result, we can deploy
and in part by the Qinlan Project of Jiangsu Province. (Corresponding author: honeypots reasonably in the AMI networks to consolidate the
Yanfei Sun.) Paper no. TSG-01182-2016.
K. Wang, M. Du, and Y. Sun are with the Jiangsu High Technology defense systems according to the equilibriums. Our proposed
Research Key Laboratory for Wireless Sensor Networks, Nanjing University model does not only improve the detection rate but also helps
of Posts and Telecommunications, Nanjing 210003, China (e-mail: reduce energy consumption.
kwang@njupt.edu.cn; dumiao0118@163.com; sunyanfei@njupt.edu.cn).
S. Maharjan is with Simula Research Laboratory, 1325 Fornebu, Norway, To this end, our main contributions are listed as follows:
and also with the University of Oslo, 1325 Oslo, Norway (e-mail: • We introduce the idea of deploying honeypots into
sabita@simula.no). an AMI network for designing secure communica-
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. tions between the operators and the consumers in the
Digital Object Identifier 10.1109/TSG.2017.2670144 smart grid.
1949-3053 c 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
WANG et al.: STRATEGIC HONEYPOT GAME MODEL FOR DDoS ATTACKS IN SMART GRID 2475
The payoff obtained with legitimate users’ strategy 2 can be Algorithm 1: Optimal Strategies for Honeypot Game
computed as JW1 (2 ) = P(Z1 | 1 ) ∗ 0 + P(Z2 | 1 ) ∗ 0 + Model
P(Z3 | 1 ) ∗ 0 = 0. Input: θ , ϕ,
, ω, λ, δ and ς
The payoff of attackers with strategy (1 ) is Output: Optimal strategies {(ii , jj ), (ii , jj )}
/* Initialize the strategies, {i , j } */
JW2 (1 ) = P(Z1 | 1 ) ∗ (−δς ) + P(Z2 | 1 ) ∗ ϕλ /* Find the stable state */
+ P(Z3 | 1 ) ∗ σ ς if θ < ϕ/2 + ϕ then
=
(δς + ϕλ) + ω(δ + σ )ς − δς. (4) if ω < (ϕλ − δς )/2(δ + σ ) ∧
< 1/2 then
choose optimal strategy {(1 , 1 ), (1 , 1 )}.
Similarly the payoff of attackers with strategy (2 ) can be end
computed as JW2 (2 ) = P(Z1 | 1 ) ∗ 0 + P(Z2 | 1 ) ∗ 0 + else
P(Z3 | 1 ) ∗ 0 = 0. cannot achieve a BNE.
Theorem 1: A BNE strategy {(1 , 1 ), (1 , 1 )} exists in end
the honeypot game model provided if ω > (ϕλ − δς )/2(δ + σ ) ∧
< 1/2 then
choose optimal strategy {(1 , 1 ), (1 , 2 )}.
ϕ 1 ϕλ − δς end
θ< ,
< , ω< .
2+ϕ 2 2(δ + σ ) else
cannot achieve a BNE.
Proof: We first assume that JZ1 (1 ) = JZ1 (2 ). Then, we
end
have
end
ϕ
θ= . (5) else
2+ϕ if ω > (ϕλ − δς )/2(δ + σ ) ∧
> 1/2 then
From the perspective of the service-side, according to (5), choose optimal strategy {(2 , 1 ), (2 , 2 )}.;
when the visitors are attackers if θ < ϕ/(2 + ϕ), 1 would end
be the dominant strategy for the SP. In that case, the SP will else
provide the real service. Otherwise, if θ > ϕ/(2 + ϕ), 2 cannot achieve a BNE.
would be the dominant strategy. Considering that the players end
in this game should choose the dominant strategies, we can end
obtain the dominant strategy {(1 , 1 )} for visitors, which is
their strategy {(1 , 1 )} under the condition θ < ϕ/(2 + ϕ),
if θ > ϕ/(2 + ϕ), the dominant strategy is {(2 , 1 )}. Equilibrium (BNE) strategy {(1 , 1 ), (1 , 1 )} for the game
We explain and prove the dominant strategy of the SP when when (10) is true.
the visitors use strategy {(1 , 1 )}. Then, we need to evaluate When θ > ϕ/(2 + ϕ), the dominant strategy of services
whether the strategy {(1 , 1 )} is the dominant strategy or not is {(2 , 1 )}. In this case, the strategy {(1 , 1 ), (1 , 1 )}
from the perspective of the visitors. Assuming that JW1 (1 ) = cannot result a BNE in the game according to (7), (8),
JW1 (2 ) and JW2 (1 ) = JW2 (2 ), we have (9) and (10).
1 Analogously, two other BNE strategies
= (6) {(1 , 1 ), (1 , 2 )} and {(2 , 1 ), (2 , 2 )} exist in
2
δς − ω(δ + σ ) the game under conditions (11) and (12), respectively.
= , (7)
ϕλ + δς ϕ 1 ϕλ − δς
θ< ,
< , ω> . (11)
Solving (6) and (7) simultaneously, we obtain 2+ϕ 2 2(δ + σ )
ϕ 1 ϕλ − δς
δς − ω(δ + σ ) 1 θ> ,
> , ω> . (12)
= , (8) 2+ϕ 2 2(δ + σ )
ϕλ + δς 2
In the next section, we will analyze the optimal strategies
which further yields for legitimate users and attackers according to the BNEs.
ϕλ − δς
ω= , (9)
2(δ + σ ) III. O PTIMAL S TRATEGIES
ϕ 1 ϕλ − δς We first analyze the BNEs in the honeypot game model
θ < ,
< , ω< . (10)
2+ϕ 2 2(δ + σ ) compared to the traditional game in terms of equilibrium
strategies. We, then, analyse the payoffs of legitimate users
Consider the case when θ < ϕ/(2+ϕ). In this case, accord-
and attackers via game trees.
ing to (6), if the probability of honeypot service is
< 1/2,
the legitimate users’ strategy 1 will be the dominant strat-
egy for SP’s strategy {(1 , 1 )}. Similarly, according to (7), A. Nash Equilibrium Analysis in Honeypot Game Model
if the attacker uses strategy 1 , the dominant strategy for The honeypot game model is considerably different than
the SP will be {(1 , 1 )} when ω < (ϕλ − δς )/2(δ + σ ). the traditional game model in terms of the equilibrium condi-
Thus, from (8) and (9), we can obtain a Bayesian-Nash tions. In a traditional Bayesian game, strategy {1 , (1 , 2 )}
2478 IEEE TRANSACTIONS ON SMART GRID, VOL. 8, NO. 5, SEPTEMBER 2017
TABLE II
S IMULATION S ETTINGS
Similarly, while the energy consumption is slightly less than A. Security Issues in AMI
in the AM model, the detection rate is close to what the AM Security issues for AMI in the smart grid has been widely
model can provide. When {
, ω} = {(0.6, 0.2)}, the energy studied. Some work focused on intrusion detection. For
consumption is close to what the AM model incurs, but the instance, Faisal et al. [11] presented an Intrusion Detection
detection rate is higher than in the AM model. When we con- System (IDS) architecture using data stream for AMI in
tinue to increase
and decrease ω, which means we need the smart grid, and analysed the performance of exist-
to deploy more honeypots in the networks. However, we find ing data stream mining algorithms with an IDS data set.
that the energy consumption and the detection rate of these two Wang et al. [12] introduced a framework of cost-model for
performances are worse than in the AM model. In other words, evaluating the architectures of IDS. Pivotal management to
along with the change in {
, ω}, the energy consumption strengthen the smart grid security includes a significant amount
and the detection rate are also changing. By varying {
, ω}, of work. For instance, Ye et al. [13] presented a novel protocol
we can find the appropriate value to obtain dynamic balance called Integrated Authentication and Confidentiality (IAC) to
between energy consumption and detection rate. As a result, ensure the security in AMI communication. Liu et al. [14]
we can conclude that presented a novel management model for large number of
1)In this testbed, {
, ω} = {(0.6, 0.2)} can achieve the devices.
optimal performance in terms of energy consumption and In addition, we can list several possible threats related to
detection rate. Therefore, we can deploy about 10 honeypots the development of smart grid as follows: a) high complex-
and 3 anti-honeypots in this AMI network testbed. ity is likely to cause the network to be much easily attacked,
2)In a dynamic network, more honeypots deployed in the as well as lead to some unknown errors. b) some new types
network do not necessarily mean that the defense performance of attacks are emerged due to the interactions between dif-
is more effective. ferent networks, and further cause the collapse of the defense
3)When the performance of energy consumption and system. c) multiple interfaces in the network may increase
detection rate reaches a dynamic balance (e.g., {
, ω} = the possibility of a DDoS attack. d) multiple nodes in the
{(0.55, 0.25)}, the system model will achieve the optimal network are potential threats, since they are very vulnerable
strategies for both the attackers and the defenders. to the attackers. e) a large number of data collection and two-
way transmission may cause the consumer privacy and data
V. R ELATED W ORK confidentiality to be attacked [15].
In this section, we propose a brief summary of the state
of the art literature on security issues in AMI, honeypot for B. Honeypot for DDoS
DDoS attacks, and the use of game theory for modeling DDoS Honeypot is one of the security resources, which is used as
attacks. a trap to lure the attacker. The concept of honeypots has long
WANG et al.: STRATEGIC HONEYPOT GAME MODEL FOR DDoS ATTACKS IN SMART GRID 2481
been used to improve security in different systems [16]–[18]. of attackers. We presented a honeypot game strategy to ana-
Provos [19] presented ‘honeyd’, which is a honeypot software lyze the strategic interactions between the attackers and the
package to monitor large-scale honeynet. Dagon et al. [20] defenders. Simulation results showed that the energy consump-
presented the ‘honeyStat’ system to utilize honeypots to detect tion and the detection rate can be improved with the proposed
worm attacks in the networks. Jiang and Xu [21] presented model, which indicate that the honeypot game strategy can be
a virtual honeynet system with a distributed presence and applied to an AMI network to protect the data and to further
centralized operation. Wang et al. [22] presented a hybrid ensure the security of AMI networks in the smart grid.
and distributed honeypot architecture to capture attack traffic.
Vrable et al. [23] designed large-scale honeynet systems to R EFERENCES
obtain high-fidelity attack data. Tang and Chen [24] presented
[1] F. Ye, Y. Qian, and R. Q. Hu, “A real-time information based demand-
a novel ’double-honeypot’ detection system which can effec- side management system in smart grid,” IEEE Trans. Parallel Distrib.
tively detect worm attacks. Syst., vol. 27, no. 2, pp. 329–339, Feb. 2016.
Some previous studies pointed out the idea that honey- [2] Y. Zhang et al., “Securing vehicle-to-grid communications in the smart
grid,” IEEE Wireless Commun., vol. 20, no. 6, pp. 66–73, Dec. 2013.
pots can be deployed in the smart grid to attract, detect, [3] K. Wang et al., “A survey on energy Internet: Architecture,
and gather attack information [25]. Through the use of approach, and emerging technologies,” IEEE Syst. J., to be published,
Virtual Manufacturing (VM) monitors, honeypots are rea- doi: 10.1109/JSYST.2016.2639820.
[4] N. Krawetz, “Anti-honeypot technology,” IEEE Security Privacy, vol. 2,
sonably deployed in the network [24], in order to monitor no. 1, pp. 76–79, Jan./Feb. 2004.
attacker activities [3], [26], [27]. Hastings et al. [28] set a [5] S. Maharjan, Q. Zhu, Y. Zhang, S. Gjessing, and T. Başar, “Demand
low-interaction honeypot in the smart grid and recorded the response management in the smart grid in a large population regime,”
IEEE Trans. Smart Grid, vol. 7, no. 1, pp. 189–199, Jan. 2016.
attack data for 6 months. Shadow honeypot [29] is a new [6] K. Wang, X. Qi, L. Shu, D.-J. Deng, and J. J. P. C. Rodrigues,
hybrid detection method, which verifies the abnormal prog- “Toward trustworthy crowdsourcing in the social Internet of Things,”
nosis and improves the algorithm of hybrid detection via IEEE Wireless Commun., vol. 23, no. 5, pp. 30–36, Oct. 2016.
[7] K. Wang, Y. Shao, L. Shu, C. Zhu, and Y. Zhang, “Mobile big data
feedback mechanism. fault-tolerant processing for eHealth networks,” IEEE Netw., vol. 30,
no. 1, pp. 36–42, Jan./Feb. 2016.
[8] Y. Zhang et al., “Cognitive machine-to-machine communications:
C. Game Theory for Modeling DDoS Visions and potentials for the smart grid,” IEEE Netw., vol. 26, no. 3,
Game theory has been widely used to analyze the secu- pp. 6–13, May/Jun. 2012.
[9] K. Wang et al., “Game-theory-based active defense for intrusion detec-
rity of critical systems. Mirkovic and Reiher [30] presented tion in cyber-physical embedded systems,” ACM Trans. Embedded
a classification of DDoS attacks and defense mechanisms. Comput. Syst., vol. 16, no. 1, 2016, Art. no. 18.
Peng et al. [31] proposed a review of defense mechanisms [10] H. Moosavi and F. M. Bui, “A game-theoretic framework for robust
optimal intrusion detection in wireless sensor networks,” IEEE Trans.
and DDoS attacks based on networks. Jiang et al. [32] intro- Inf. Forensics Security, vol. 9, no. 9, pp. 1367–1379, Sep. 2014.
duced a two-person zero-sum game to deal with DDoS traffic [11] M. A. Faisal, Z. Aung, J. R. Williams, and A. Sanchez, “Data-stream-
injection. Xu and Lee [33] proposed a game-theoretic model based intrusion detection system for advanced metering infrastructure in
smart grid: A feasibility study,” IEEE Syst. J., vol. 9, no. 1, pp. 31–44,
to solve DDoS attacks and analyzed the performance of the Mar. 2015.
defense system. Yan and Eidenbenz [34] presented a novel [12] K. Wang, Z. Ouyang, R. Krishnan, L. Shu, and L. He, “A game theory-
mechanism, providing ISPs to address DDoS attacks in a based energy management system using price elasticity for smart grids,”
IEEE Trans. Ind. Informat., vol. 11, no. 6, pp. 1607–1616, Dec. 2015.
non-cooperative game. Mohi et al. [35] proposed a Bayesian
[13] F. Ye, Y. Qian, and R. Q. Hu, “A security protocol for advanced metering
game model to defend against DDoS attacks in wireless sen- infrastructure in smart grid,” in Proc. IEEE Globecom, Austin, TX, USA,
sor networks. Zang et al. [36] utilized a Bayesian game model 2014, pp. 649–654.
to deal with the DDoS attack. Chai et al. [37] proposed the [14] N. Liu, J. Chen, L. Zhu, J. Zhang, and Y. He, “A key management
scheme for secure communications of advanced metering infrastruc-
game model in a continuous setting, and the Nash equilibrium ture in smart grid,” IEEE Trans. Ind. Electron., vol. 60, no. 10,
can be computed to address attack detection problems. pp. 4746–4756, Oct. 2013.
Nevertheless, to the best of our knowledge, there is lit- [15] R. Yu et al., “Cognitive radio based hierarchical communications
infrastructure for smart grid,” IEEE Netw., vol. 25, no. 5, pp. 6–14,
tle work towards the deployment of honeypots for enhancing Sep./Oct. 2011.
security in the smart grid, particularly, focusing the analyt- [16] K. Wang, L. Yuan, T. Miyazhaki, S. Guo, and Y. Sun, “Anti-
ical models. Our proposed model can detect DDoS attacks eavesdropping with selfish jamming in wireless networks: A Bertrand
game approach,” IEEE Trans. Veh. Technol., to be published,
by deploying honeypots in AMI networks with higher proba- doi: 10.1109/TVT.2016.2639827.
bility than traditional methods. We utilize a game theoretical [17] C. K. Dimitriadis, “Improving mobile core network security with hon-
approach to analyze and prove the added security level due to eynets,” IEEE Security Privacy, vol. 5, no. 4, pp. 40–47, Jul./Aug. 2007.
[18] K. Wang and M. Wu, “Nash equilibrium of node cooperation based on
the honeypots while also capturing characteristic features of metamodel for MANETs,” J. Inf. Sci. Eng., vol. 28, no. 2, pp. 317–333,
the attackers, legitimate users, and the service providers. Thus, 2012.
our model is expected to be useful in deploying honeypots in [19] N. Provos, “A virtual honeypot framework,” in Proc. 13th USENIX
Security Symp., San Diego, CA, USA, 2004, p. 1.
a real AMI scenario. [20] D. Dagon et al., “HoneyStat: Local worm detection using honeypots,”
in Proc. 7th Int. Symp. RAID, 2004, pp. 39–58.
[21] X. Jiang and D. Xu, “Collapsar: A VM-based architecture for network
VI. C ONCLUSION attack detention center,” in Proc. 13th USENIX Security Symp.,
In this paper, we introduced honeypots into the AMI San Diego, CA, USA, 2004, p. 2.
[22] K. Wang, M. Du, Y. Sun, A. Vinel, and Y. Zhang, “Attack detection
network in the smart grid to address DDoS attacks. In addition, and distributed forensics in machine-to-machine networks,” IEEE Netw.,
we considered the anti-honeypot problem from the perspective vol. 30, no. 6, pp. 49–55, Nov./Dec. 2016.
2482 IEEE TRANSACTIONS ON SMART GRID, VOL. 8, NO. 5, SEPTEMBER 2017
[23] M. Vrable et al., “Scalability, fidelity and containment in the Potemkin Miao Du is currently pursuing the postgraduation
virtual honeyfarm,” in Proc. ACM Symp. SOSP, Brighton, U.K., 2005, degree in information network with the Nanjing
pp. 148–162. University of Posts and Telecommunications, China.
[24] Y. Tang and S. Chen, “Defending against Internet worms: A signature- His current research interests include wireless sen-
based approach,” in Proc. IEEE INFOCOM, Miami, FL, USA, 2005, sor network, social networks, security, game the-
pp. 1384–1394. ory, smart grid communications, and cyber-physical
[25] T. Flick and J. Morehouse, Securing the Smart Grid: Next Generation systems.
Power Grid Security. Rockland, MA, USA: Syngress, 2010.
[26] H. A. Lagar-Cavilla et al., “SnowFlock: Rapid virtual machine cloning
for cloud computing,” in Proc. ACM Eur. Conf. Comput. Syst.,
Nuremberg, Germany, 2009, pp. 1–12.
[27] N. Provos and T. Holz, Virtual Honeypots: From Botnet Tracking to
Intrusion Detection. Upper Saddle River, NJ, USA: Addison-Wesley,
2007, pp. 201–211.
[28] J. Hastings, D. M. Laverty, and D. J. Morrow, “Tracking smart grid hack-
ers,” in Proc. 49th Int. Univ. Power Eng. Conf. (UPEC), Cluj-Napoca,
Romania, 2014, pp. 1–5.
[29] K. G. Anagnostakis et al., “Shadow honeypots,” Int. J. Comput. Netw.
Security, vol. 2, no. 9, pp. 1–15, 2010.
[30] J. Mirkovic and P. Reiher, “A taxonomy of DDoS attack and DDoS
defense mechanisms,” in Proc. ACM SIGCOM, Portland, OR, USA,
2004, pp. 39–53.
[31] T. Peng, C. Leckie, and K. Ramamohanarao, “Survey of network-based
defense mechanisms countering the DoS and DDoS problems,” ACM
Comput. Surveys, vol. 39, no. 1, pp. 60–67, 2007.
[32] H. Jiang, K. Wang, Y. Wang, M. Gao, and Y. Zhang, “Energy big data:
A survey,” IEEE Access, vol. 4, pp. 3844–3861, 2016.
[33] J. Xu and W. Lee, “Sustaining availability of Web services under dis- Sabita Maharjan received the M.Eng. degree
tributed denial of service attacks,” IEEE Trans. Comput., vol. 52, no. 2, in wireless communication from the Antenna
pp. 195–208, Feb. 2003. and Propagation Laboratory, Tokyo Institute of
[34] G. Yan and S. Eidenbenz, “DDoS mitigation in non-cooperative envi- Technology, Tokyo, Japan, in 2008, and the Ph.D.
ronments,” in Proc. Int. Conf. Netw., Singapore, 2008, pp. 599–611. degree in network and distributed systems from the
[35] M. Mohi, A. Movaghar, and P. M. Zadeh, “A Bayesian game approach University of Oslo, Oslo, Norway, and the Simula
for preventing DoS attacks in wireless sensor networks,” in Proc. Int. Research Laboratory, Fornebu, Norway, in 2013,
Conf. Commun. Mobile Comput., Kunming, China, 2009, pp. 507–511. where she is currently a Post-Doctoral Fellow. Her
[36] W. Zang, P. Liu, and M. Yu, “How resilient is the Internet against DDoS current research interests include wireless networks,
attacks?—A game theoretic analysis of signature-based rate limiting,” network optimization, security, game theory, smart
Int. J. Intell. Control Syst., vol. 12, no. 4, pp. 307–316, 2007. grid communications, and cyber-physical systems.
[37] B. Chai, J. Chen, Z. Yang, and Y. Zhang, “Demand response manage-
ment with multiple utility companies: A two-level game approach,” IEEE
Trans. Smart Grid, vol. 5, no. 2, pp. 722–731, Mar. 2014.