Aranya Chakrabortty. North Carolina State University. Department of Electrical and Computer Engineering. March 23, 2017

Transcription

1 Cyber-Physical Implementation of Wide-Area Control using ExoGENI-WAMS Testbed Aranya Chakrabortty Department of Electrical and Computer Engineering North Carolina State University March 23, 2017

2 From Centralized to Distributed Architecture Centralized WAMS Distributed WAMS 1. Formulation of centralized strategies to distributed algorithms 2. Investigation of cyber-physical issues: delays and cyber-security 3. Validation using cyber-physical networked cloud computing testbed 2

3 Distributed Asynchronous Algorithms Area 1 Area 3 PMU Central PDC at ISO PMU y 11 (t) y 12 (t) y 31 (t) y 32 (t) z z * d 2 * d 2 ( a, w ) k+ 1 k 1 1 ( a, w ) k+ 1 k 2 2 z * d 1 ( a, w ) z k+ 1 k 3 3 ( a, w ) k+ 1 k 4 4 * d 2 * d 2 Assumption of S-ADMM: the communication between local PDCs and central PDC is completely synchronized. Practically, it is not always possible. y 21 (t) y 22 (t) y 41 (t) y 42 (t) PMU PMU Area 2 Area 4 Proposed method: to counteract asynchrony by defining a set of flexible deadlines for message arrival in every PDC, and by modifying the update rules based on these deadlines. Our Objective: Study the impact of asynchronous communication on the convergence of distributed ADMM-based algorithms. Result algorithm is called A-ADMM. Approach: Step 1. Consider a probabilistic traffic model for modeling delays in Internet Step 2. Propose different update strategies to immune asynchrony Step 3. Analyze the convergence of the A-ADMM algorithm on 1) delay distribution 3 parameters 2) different update strategies

4 Delay Model for Wide-Area Communication 1. Minimum deterministic delay, 2. Internet traffic delay (alternating renewal process) 3. Router processing Delay (Gaussian process) Then, the total Probability Density Function (PDF) We derive the Cumulative Distribution Function (CDF) µ (1 ) ( ) 2 ( ) [erf( ) + erf( t µ )] p λ σ + µλ λσ + µ Pt e [erf( ) + erf( t λσ µ = + )] 2 2σ 2σ N 2σ 2σ Parameters: 1. - mean deterministic delay (5.3ms), 2. - derivative value of delay of router processing(0.078ms), 3. - mean length of the closure period (1/1.39), 4. - probability of open period of the path with no Internet traffic(0.58), 5. - cumulative distribution function of d*, d* is the delay threshold. 4

5 Security Enhancement of Distributed Optimization Question: Cyber-physical architectures are prone to failures and attacks on their physical infrastructure, as well as cyber attacks on their data management and communication layer. Objective: To investigate how distributing a monitoring functionality over multiple estimators can guarantee significantly more resiliency against extreme events. Attack Scenarios: 1. Malware attack that disrupts the normal execution of monitoring algorithms 2. Flooding attack by generating malicious traffic to delay messages 3. Malfunction of physical infrastructure due to natural calamities such as storms and earthquakes 5

6 Resilient Strategy for Centralized RLS

7 Resilient Strategies for Distributed S-ADMM Redundancy-based Strategy Local PDC runs as dual roles of local estimator and central estimator 7

8 Experimental Validation on Federated Testbeds Federated Testbeds: RTDS Lab of NC State + DETER Lab of Univ. of South California Centralized Architecture 1 local PDC attacked 2 local PDCs attacked 3 local PDCs attacked 8

9 Experimental Validation on Federated Testbeds Distributed Architecture 9

10 Project Demos: DETER Demo at Smart-America 2014 Best Energy App Award at US Ignite 2015 US Ignite & NIST Smart Cities Application Summit, Austin, TX, 2016

11 ExoGENI-WAMS Testbed Bring Concepts of Cloud Computing and Software Defined Networking into Research of Wide-Area Monitoring and Control with PMU data Wide-Area Monitoring and Control is a typical cyber-physical system Problems of the physical subsystem 1. Accessing of real PMU measurements due to privacy and non-disclosure issues 2. Not sufficient for studying dynamics of the entire system due to limited coverage Requirements of the cyber subsystem To utilize next-generation cyber-infrastructure technologies: 1. high-speed virtual networking 2. high performance networked cloud computing 3. virtualization and data management Objective: build up a perfect cyber-physical testbed for WAMS research Result: ExoGENI-WAMS Testbed Physical subsystem Hardware-In-Loop Framework (RTDS + PMU-based WAMS) Cyber subsystem Networked Cloud Computing Platform (ExoGENI) 11

12 Architecture of ExoGENI-WAMS Testbed US-wide ExoGENI at RENCI/UNC Chapel Hill PMU based WAMS at NC State

13 Components: RTDS-PMU based WAMS RTDS two racks, 50 us of time step, RSCAD software to develop models for the RTDS to simulate GATO hardware interface of Gigabit Transceiver Analog Output to generate voltage and current waveforms to the PMUs GTNETx2 Gigabit Transceiver Network interface card to communicate with remote station. Multiple protocols (TCP socket, DNP, ) IEEE 754 floating-point and integer type. PMU 5 units: 3 SEL-421 & 2 SEL-487 Functions: accepting IRIG-B signal for satellite synchronization RTDS GTNETx GPS SEL-421 Switch ExoGENI GPS SEL-2407 Satellite-Synchronized Clock SEL

14 Networked Cloud Computing Testbed-ExoGENI ExoGENI provides in virtual IaaS services for innovative research on distributed applications for Wide-Area Monitoring and Control (14 rack sites at universities & labs over the US) Software Layer Other GENI Resources GENI Experimenter Tools (Flukes) Other GENI Resources RENCI SM ExoSM (Slice Manager) N SM RENCI AM LEARN AM BEN AM I2 AM NLR AM SL AM ANI AM N AM Physical Layer: LEARN OpenFlo w RENCI Rack BEN X I2/ION PMU based WAMS at NC State NLR X Circuit Providers StarLight GENI AM API ORCA AM API Native IaaS API ESnet OpenFlo w Rack N

15 Validations of ExoGENI-WAMS testbed Visualization of Power Grid Delay Evaluation of CLS, DLS and RLS Distributed Oscillation Monitoring Algorithm Distributed Storage System (DSS) for Multiple Applications Distributed Control Algorithm

16 Case Study I -Distributed Storage System with S-ADMM Synchronized ADMM + Storage System Step 1: PMUs keep storing data into Storage System PDC1 PDC2 PDC4 WAMS Apps Layer Storage Layer Distributed Real-time Storage System PMU Layer

17 Case Study II - Distributed Control Algorithm US West Coast Approximate Generation Capacities Up to 3000 MW Up to 6000 MW Up to 9000 MW Above MW Close the loop from cloud to grid E 2 δ 2 E 1 δ 1 r 12 jx 12 V 1 θ 1 Area 2 ASG 3 E 3 δ 3 jx 3 P 3 V 3 θ MW V 5 θ MW Area 1 V 4 θ 4 E 5 δ 5 E 4 δ MW Area 3 VM VM VM Area MW 9000 MW VM VM Area 5 Third-Party Private Cloud + Controllable Network

18 Implementation of Distributed Control Algorithm Control Signals from ExoGENI Control Signals from RSCAD sec P G1Input units t t P G2Input units P t1-t2 = 200 millseconds! G3Input units P G4Input units P Control delays using SDN traffic control-rules in ExoGENI G5Input units

19 Comparison of LQR Controller Performance Performance of RSCAD-based LQR-WAC compared against a cloudcomputing implementation using the ExoGENI Network. 19

20 Conclusions and Future Work Conclusions Develop distributed delay-robust algorithms for wide-area oscillation mode monitoring of power systems Investigate the convergence performance of these distributed algorithms on delay distribution parameters and different variants of asynchronous strategies ExoGENI-WAMS-DETER testbed Validations of these distributed architecture using distributed cloud computing Future Work Investigate the scalability problem of distributed algorithms Resilience of ExoGENI using SDN principles Delay management in ExoGENI using SDN principles 20

21 Thank You Website: 21