Distributed Control And Intelligence Using Multi Agent Systems

1Distributed Control andIntelligence UsingMulti‐Agent SystemsChen‐Ching Liu1,21Washington State University2University College Dublin

2Underfrequency Load Shedding (UFLS) Conventional UFLS schemes— In operation for decades;— Locall decisiondbyb relayslddistributedb d in various locationsl(no(communication);— Developedp byy off‐line analysis;y ;— Excessive or insufficient load shedding is not uncommon.

3Underfrequency Load Shedding (UFLS) Centralized UFLS schemes— Optimal load shedding decision can be achieved;— Relevantlinformationfmust beb transmittedd to a centrallprocessing facility;— Optimizationpcalculation is pperformed;;— Delay of a measurement may result in a slow response of theglobal computational process.

4Underfrequency Load Shedding (UFLS) Distributed UFLS schemesLet’ss talk!Let— Distributed computation with information sharing;— Adaptive for contingency scenarios and operatingconditions;— Adaptive to a change in system topology;— Amount of load to be shed may not be optimized for lack ofgglobal information;;— Effectiveness is the goal, e.g., quickly stop the frequencydecay beyond a tolerance of frequency deviation.

5Distributed UFLS Agent framework– Classifications PowerPplantl t agentt (PA) Substation agent (SA)Agent communication layer– Sensors Phasor measurement unit ((PMU)) Phasor data concentrator (PDC) Circuit breaker monitor (CBM)– Actuators RelaysR landd circuiti i breakersb kPMU 1 PDCProcessorPMU nACTUATORSCBM 1 CBM mL dLoadsheddingrelaysConcentratorunitSENSORSLog files for storageof event recordsAgent framework

6Distributed UFLS

7Distributed UFLS Multi‐agent based UFLS scheme—Distributingg stepp Distributed load shedding amounts should bedetermined quickly;qy; Load buses closer to generators are more effective; Bus voltage magnitudes and angles are acquired fromPMUs locally, and shared by reaching agreementamong agents.

8Reaching Agreement Sharing information among agents– Consensus problemp Each agent has an initial state and all agents must agree onthe same value in the final state (not necessarily the originalinitial state).– Average‐consensus problem Each agent has an initial state and all agents agree on theaverage of their initial states; If ththe iinitialiti l statet t off an agentt iis a vector,t ththe average off thethcorresponding elements among all vectors is reached and astate with the same dimension is obtained.

9Reaching Agreement

10Reaching Agreement

11Reaching Agreement

12Distributed UFLSAgentsPA1PA2PA3PA4PA5Buses2, 306, 3125, 3729, 3839Initial Values [‐16.59, 0, 0, 0, 0] [0, ‐63.37, 0, 0, 0] [0, 0, ‐23.71, 0, 0] [0, 0, 0, ‐78.72, 0] [0, 0, 0, 0, ‐425.34]Activee Power Imballance (MW)0-20 PPA1 PPA2-40 PPA3 PPA4-60 PPA5-80-100-12012005101520Number of Iterations2530Information sharing process of five PAs (from the perspective of PA1)

13Distributed UFLS Comparison – “Advanced” load sheddinggystrategyRelay GroupLoad Shedding in Each StepN b off StepsNumberSFrequency Threshold (Hz)Frequency DerivativeThreshold ((Hz/s)/)Time of FrequencyMeasurement (s)Time to Open the CircuitBreaker (s)G1 – 1st Group2.5%1059.8, 59.72, 59.64, , 59.08G2 – 2nd Group1.6%1659.0, 58.9, 58.8, , 57.5[‐1.0, ‐0.3]NULL0.10.10.0750.075

14Distributed UFLS Contingency scenarioTime(sec)EventActive Powerfrom SIto NI (MW)320.480.5Tripping of line 6‐111.5Tripping of line 4‐14264.052.2Tripping of line 16‐17238.96Total active power deficiency in NI:823.49 MWPower grid splitting

15Distributed UFLS ComparisonLoad SShedding PPercentage100.00%Advanced Load Sheddingg Method90 00%90.00%80.00%MAS-Based Distributed Method70.00%60 00%60.00%50.00%40.00%30.00%20.00%10.00%0.00%Bus No. 34781825262728Comparison of load shedding distribution293139

16Distributed UFLS ComparisonMethodsFrequency DecayEnds (sec)Active PowerLoad (MW)Reactive PowerLoad (MVAR)Frequency (Hz)Multi‐agentgAdvanced baseddistributed84.5643.35607.73162 07162.0783 7983.7959.859.86643.35 - 607.73 35.62 MW162.07 - 83.79 78.28 MWSSystemresponses (no( secondary control))

17Agent Based Modeling: Market Rules EvaluationAgent‐Based Evaluation of market rules– Complexitypy of the market structure Strategic interaction between participants; Underlying physics.– Difficult to evaluate implications of potentialchanges to market rules;– Day‐ahead market (DAM) is modeled as a MAS;– Each participant is modeled as an agent.

18Agent Based Modeling: Market Rules EvaluationAgent‐Based System structure and messageqflowingg sequence

19Architecture of MASs Strategic power infrastructureydefense ((SPID)) system– Hierarchical, layered multi‐agent system concept;– Hybrid multi‐agent systemmodel.

20Distributed Control Distributed control systems (DCSs)– Control units are distributed throughoutthe system;– Large, complex industrial processes,geographically distributed applications;– Utilize distributed resources forcomputation with information sharing;– Adapt to contingency scenarios andoperating conditions;– Self‐adapt to a change in system topology.

21Distributed Intelligence Distributed artificial intelligenceg(DAI)– Used by distributed solutions for complexproblems that require intelligence;– Based on different technologies, e.g., distributedexpert systems, planning systems, or blackboardsystems;– Closely related to the field of MASs;– Consisting of autonomous learning agents;– Agentsgare often heterogeneous.g– Example: VOLTTRON by Pacific NorthwestNational Lab

22For Further Information C.‐C.CC Liu,Li J.J Jung,JG T.G.T Heydt,H d V.V Vittal,Vi l andd A.A G.G Phadke,Ph dk “The“Th strategici power infrastructurei fdefense (SPID) system. A conceptual design,” IEEE Control Systems Magazine, vol. 20, no. 4,pp. 40–52, Aug. 2000. H. Li, G. W. Rosenwald, J. Jung, and C.‐C. Liu, “Strategic power infrastructure defense,”Proceedings of the IEEE, vol. 93, no. 5, pp. 918–933, May 2005. N. Yu, C.‐C. Liu, and J. Price, "Evaluation of market rules using a multi‐agent systemmethod," IEEE Transactions on Power Systems, Vol. 25, pp. 470‐479, Feb. 2010. J. Xie, C.‐C. Liu, and M. Sforna, “Distributed underfrequency load shedding using a multi‐agent system,” Proc. IEEE PowerTech (POWERTECH ’15), Eindhoven, Netherlands, Jul. 2015. J. Xie, C.‐C. Liu, M. Sforna, M. Bilek, and R. Hamza, “Intelligent physical security monitoringsystem for power substations,” Proc. 12th Intell. Syst. Appl. Power Syst. (ISAP ’15), Porto,Portugal, Sep. 2015.

Distributed Control 20 Distributed control systems (DCSs) - Control units are distributed throughout the system; - Large, complex industrial processes, geographically distributed applications; - Utilize distributed resources for computation with information sharing; - Adapt to contingency scenarios and