Y9.ET1.3 Implementtion of Secure Energy ngement ginst Cyber/physicl Attcks for FREED System Project Leder: Fculty: Students: Dr. Bruce cillin Dr. o-yuen Chow Jie Dun 1. Project Gols Develop resilient cyber-physicl control strtegy for FREED system to secure the energy scheduling nd physicl opertion: Extending to cyber-physicl security: Explore possible ttcks on both energy scheduling nd physicl opertion, nlyze the impcts in terms of economic benefit nd system stbility. Interction between cyber/physicl lyer detection: Investigte the correltion between the cyber lyer behvior nd physicl lyer response, inconsistent behviors in both lyer re helpful to disclose the misconduct devices. Collbortive cyber-physicl countermesures: Integrting the cyber lyer resilience with physicl lyer secure lgorithms to build resilient cyber-physicl control frmework ginst possible ttcks. Implement the resilient cyber-physicl control frmework in DGI pltform nd further test the lgorithm in HIL nd GEH testbeds. 2. Role in Support of Strtegic Pln This project closes the control loop for the distributed energy mngement to remin optiml in the presence of cyber ttcks. oreover, this project provides tngible demonstrtion of how the coopertive distributed energy mngement should be implemented in HIL/GEH testbed. 3. Fundmentl Reserch, Technologicl Brriers nd ethodologies The technology brrier is how to build distributed monitoring system to fit for the distributed control frmework. The methodology to del with it is to develop neighborhood-wtch mechnism in which ech node is responsible for monitoring its neighbors. A reputtion index is introduced to reflect the credibility of the neighbors, if one neighbor is continuing sending out flse informtion, the reputtion index will indicte tht it is compromised. 4. Achievements 4.1 New Dt Integrity Attck on CoDES lgorithm The CoDES lgorithm [1] is fully distributed optimiztion method, in which the DGI nodes communicte with only neighbors nd determine the locl genertion schedule. It brings some significnt dvntges to the system, in terms of sclbility nd robustness [2]. However, the fully distributed frmework is lso vulnerble to mlicious cyber ttcks, in which some devices might choose not to collborte with neighbors, but to seek for selfish objectives [3]. For exmple, Fig.1 shows selfish DESD (DESD 1) in the FREED GEH system who wnts to mximize its own economic benefit:
Figure 1: A mlicious DESD in FREED system Step 1: Determining the most profitble schedule for itself The selfish objective of the mlicious DESD is given s: T mx p( t) P ( t) t t1 st.. t 1,..., T, x E P ( s) t x, full t s1, min, mx t 1,..., T, P P ( t) P. B B where P t denotes the power commnd to the mlicious storge device t time step t, positive vlue mens dischrging commnd; pt is the energy price t time step t; E i,fu is the storge cpcity nd x i is initil vlue of the stored energy. P i,bi nd P i,bx re the minimum nd mximum power limits of the storge device. Step 2: nipulting the power imblnce estimtion In the CoDES lgorithm, ll the devices estimte the system power imblnce in collective sense, using consensus network. Thus, ttcker could use flse locl power blnce estimtion to misled the system to overestimte or underestimte the system power imblnce. Assume ttcker sends out flse locl power imblnce estimtion P,f t in itertion k, nd the neighbors of device use P,f t for their locl updte, while the ttcker still updtes using the correct informtion P t. In this cse, due to the flse informtion P,f t, from itertion k + 1: ˆk1 k1 ˆk, f ( ) ( ) ( ) ˆk P t P t w P t P ( t), i i sys j jn which mens the locl estimtion of system power imblnce devites from the ctul vlue. When ttcker is lunching the dt integrity ttck, it keeps the ctul scheduling commnd P t to be, ppering not to be contributing to the microgrid. At the sme time, it mnipultes the devition to be exctly the sme s P. The impct of this ttck is illustrted in Fig.2.
Electricity Price High Price Power Imblnce Estimtion P ( T ) 1 P ( T ) 2 Overestimtion Underestimtion Low price Actul Lod Actul Lod t = T1 t = T2 Figure 2: nipultion of power imblnce estimtion When the electricity price is low (t = T, the system is misled to overestimte the system lod, with difference s P T. In contrst, when the electricity price is high (t = T, the system underestimtes the system lod, with difference s P T. Consequently, while the norml devices djust their power genertions to support the flse lod, nd ttcker chrges excess power when the electricity price is low nd dischrges when the electricity price is high. With the flse lod estimtions, the norml devices hve to djust their schedule in order to meet the flse lod, while DESD 1 chrges nd dischrges ccording to the mlicious schedule P t. We illustrte the impct of the dt integrity ttck by using the FREED system locted t North Crolin Stte University, the detiled simultion setup could be found in [1]. The comprison between the norml schedule nd the ttcked one is given in Fig.3, where the green brs denote the schedule in the norml condition, nd the red brs denote the schedule under dt integrity ttck. 25 2 15 1 5 5 Grid DESD2 5-5 4 2 DESD1 (licious) DESD3-2 -5-4 Norml Condition Under Attck Figure 3: The impct of P t on the power genertion The economic impcts of the ttck on three DESDs nd the totl electricity bill re clculted s Error! Reference source not found. nd Error! Reference source not found., respectively. The results re summrized in Tble I. As we cn see, fter the ttck, the totl electricity bill increses s the ttcked schedule is not the optiml result. In the mentime, only the ttcker DESD 1 gins more economic benefit from the ttck, while the other two DESDs mke less money compred to the norml condition.
TABLE I. THE EXTRA ONEY OBTAINED BY LAUNCHING THE DATA INTEGRITY ATTACK Benefit (cents) Totl Bill DESD 1 DESD 2 DESD 3 Norml 187.2 26.8 38.56 22.35 Attcked 28.55 34.6 35.98 17.3 Difference 21.53 7.98-2.58-5.32 Impct (%) +11% +3% -6% -23.6% 4.2 Reputtion-bsed neighborhood-wtch lgorithm We proposed reputtion-bsed neighborhood-wtch lgorithm in which every node could monitor the correctness of the shred informtion from its neighbor nd counterct the ttcks [4]. Similr concept is vilble in our previous work [5], [6]. The objective of the proposed resilient control mechnism is twofold: 1) to detect the presence of ny mnipulted informtion; including λ i nd ΔP i ; 2) to recover the optiml energy schedule from the mlicious impct of the ttck. The frmework of the reputtionbsed neighborhood-wtch lgorithm is shown s Fig.4. In the following the detils of ech steps re described. Two-hop Neighbors informtion Neighbors CoDES lgorithm Power disptch result Rel-time system opertion Step 1: Informtion Estimtion Step 2: Flse informtion Verifiction Step 4: Resilient protection Step 3: licious node identifiction Neighborhood monitoring Figure 4: The frmework of distributed neighborhood-wtch lgorithm Step 1: Bsed on two-hop neighbors shred informtion in itertionk, ech node estimtes its neighbors shred informtion in next itertion k + 1, where the estimted informtion includes λ i nd ΔP i. Step 2: Ech node detects the flse informtion from its neighbors by compring the estimted vlue with the ctul received vlue. Step 3: Adjust the credibility of neighbors vi the Locl Reputtion Index. Identify mlicious bus if the corresponding reputtion drops below threshold. Step 4: Informtion from the mlicious bus is discrded by neighbors, nd the other norml buses use estimted informtion to continue the itertive process 4.3 Algorithm implementtion in DGI 2. frmework The CoDES lgorithm hs been implemented in Linux PC environment running Ubuntu 16.4 LTS. The progrm is ble to clculte the 24-hour chrging/dischrging schedule of DESDs of the 4-node FREED system s shown in Fig.5.
Figure 5: 4-node FREED system 4.3.1 Configurtion setup The CoDES module is registered in the DGI min function (Posixin.cpp) by dding "dd/disptchalgo.hpp" to the Posixin.cpp. The dd module hs lredy been registered with 9 seconds phse time. Detiled code re shown in the source code between line 326 to line 47. 4.3.2 Execution result exmple By executing (./ProsixBroker) the executble file one by one, the CoDES lgorithm will run nd ech terminl (represent ech DGI node) session will print out lgorithm informtion s the lgorithm executes.
Grid Power (output s 24X1 rry): DESD schedule (output s 24X1 rry): 5. Other Relevnt Work Being Conducted Within nd Outside of the ERC NSF progrm: Secure Algorithms for Cyber-Physicl Systems under Awrd Number 15561. 6. ilestones nd Deliverbles Q3 (9/3/216) In process of implementing the neighborhood-wtch lgorithm in DGI pltform to secure the energy scheduling lgorithm. Q4 (12/31/216) A working implementtion of the neighborhood-wtch lgorithm in DGI pltform, in process of exploring possible cyber/physicl ttcks on energy scheduling nd rel-time opertion. Q1 (3/31/217) Build the resilient cyber-physicl control frmework, develop the collbortive cyber/physicl countermesures ginst potentil ttcks. Q2 (6/3/217) The implementtion of the resilient cyber-physicl control frmework in DGI pltform. Deliverble for SV (4/217): A working implementtion of the resilient cyber-physicl control in DGI pltform Relted publictions nd reports. Finl Deliverble (8/217): System level demonstrtion of the cyber/physicl ttcks nd the corresponding countermesures. Relted publictions nd reports. 7. Plns for Next Five Yers Consider collude misbehving nodes in the system Adpt the detecting threshold considering the effect of communiction noise nd disturbnce Anlyze the optiml number of hops required to exchnge informtion under different ttck scenrios. 8. ember Compny Benefits
By demonstrting the ttcks nd the resilient opertion of the CoDES lgorithm, the member compny will be ble to see the potentil chllenges of the distributed technologies nd promising solution to del with the cyber ttcks. This secure technology could be used for other similr pplictions. 9. References [1] N. Rhbri-sr, Y. Zhng, nd.-y. Chow, Coopertive Distributed Scheduling for Storge Devices in icrogrids using Dynmic KKT ultipliers nd Consensus Networks, IEEE Power nd Energy Society Generl eeting, pp. 1 5, 215. [2] N. Rhbri-Asr, Y. Zhng, nd.-y. Chow, Consensus-bsed distributed scheduling for coopertive opertion of distributed energy resources nd storge devices in smrt grids, IET Genertion, Trnsmission & Distribution, vol. 1, no. 5, pp. 1268 1277, 216. [3] J. Dun nd.-y. Chow, Dt Integrity Attck on Consensus-bsed Distributed Energy ngement Algorithm, in IEEE Power nd Energy Society Generl eeting, 217, pp. 1 5 (submitted). [4] J. Dun, S. ember, W. Zeng, nd. Chow, Resilient Coopertive Distributed Energy Scheduling ginst Dt Integrity Attcks, 216-42th Annul Conference of the IEEE Industril Electronics Society (IECON), pp. 4941 4946, 216. [5] W. Zeng nd.-y. Chow, Resilient distributed control in the presence of misbehving gents in networked control systems., IEEE Trnsctions on cybernetics, vol. 44, no. 11, pp. 238 49, 214. [6] J. Dun, W. Zeng, nd.-y. Chow, Resilient Distributed DC Optiml Power Flow ginst Dt Integrity Attcks, IEEE Trnsctions on Smrt Grid, vol. 99, no. 99, pp. 1 1 (in press), 216.