Transcript Document
Distributed Asynchronous Algorithms & Software Systems For Wide-Area Monitoring of Power Systems Aranya * * Chakrabortty , North Carolina State University, CPS-1329780 * Mueller , Frank +University Rakesh + Bobba , Nitin of Illinois Urbana Champaign, and + Vaidya ++ Xin and Yufeng ++RENCI, University of North Carolina Project Goal Distributed Oscillation Monitoring Distributed Middleware To translate current state-of-art centralized processing algorithms for wide-area monitoring of large power grids using large volumes of Synchrophasor data to a completely distributed cyber-physical architecture. Problem statement: Compute power flow oscillation frequencies (eigenvalues), mode shapes (eigenvectors), damping, residue, participation factors, and mode energy of electro-mechanical swing dynamics from PMU measurements using distributed algorithms implemented via DRCP and DLAP.. Problem statement: Develop distributed middleware to support DRCP and DLAP algorithms for oscillation monitoring and transient stability. • Need recent PMU data from PDCs • Idea: Develop real-time distributed storage - Fault-tolerant network overlays - RT-DHT: real-time distributed hash table - Chord-like ring + finger pointers - multiple replicas of data faults OK - need deterministic wide -area networks Infrastructure: 1. Cloud computing 2. Software Defined Networks (SDN) Experiments: BEN, Exo-GENI, and GENI Study viability of DHT for fault resilience, bandwidth, latency in estimation Intellectual Merits: 1. Distributed oscillation monitoring 2. Distributed voltage monitoring 3. Distributed middleware 4. Fault-tolerance 5. Experimental verification using Exo-GENI network 6. Real-time testing of QoS and cyber-security Frequency Control Wide-area Damping Control & Optimization Dynamic State Estimation Algorithms & Decisions Event Detection & Analysis Modal Analysis & FFT Situational Awareness Voltage Stability & Load Modeling Application Level Transient Stability Assessment Winnipeg, MB, CA Real-time Dispatch Adaptive AGC Bismarck, ND St. Paul, MN PDC PDC Supervisory Controller PDC Grand Rapids, MI Control Center Level Phasor State Estimator Ames, IA PDC # 1 PDC # 2 PDC # 3 Blacksburg, VA Substation Level PDC # n Region 2 Region 1 Region n Region 3 Huntsville, AL PMU PMU PMU PMU Power System PMU PMU PMU Highly Stochastic Load Variations Unpredicted Disturbance Events PMU data from US Midwest Topology Variation Event Level a1s a1 s a1s a1 PMU PMU PMU PDC PDC PDC • • PDC PDC PMU PDC PDC Interarea Mode # 3 Cluster 3 Unidirectional Communication Cluster 2 s i s 2 a2 i s a2i a 2i a 2i Interarea Mode # 2 Interarea Mode # 2 Intra-cluster Virtualization Transient Stability & Voltage Stability Lyapunov based energy functions are useful metrics for assessment of large-signal transient stability of power grids E1 E2 S H [cos(op ) cos (δ ) sin(op )(op )] xe PMU PDC a2s a2 2 s a2s a2 Experimental Testbed 2 PMU PDC PDC PDC PDC • Develop distributed algorithms to compute energy functions using PMU data PDC PMU PMU Local PDC-PDC Communication Interarea Mode # 1 • • Centralized Data Processing PMU PDC • Design application specific fault-tolerance mechanisms to meet real-time needs of the DRCP and DLAP monitoring algorithms Crash failures Byzantine failures • Leverage the redundancy of sensors and the correlation among sensor data to reduce the cost of fault-tolerance Protecting a small subset of PMU data may be necessary and sufficient to detect false data injection attacks • Leverage application characteristics to design approximate or safe algorithms that can tolerate asynchrony and message loss Interarea PDC-PDC Communication • Dynamic Rate Control Problem (DRCP): - Find optimal PMU data exporting rates, and frequency of information exchange between local PDCs and inter-regional PDCs to minimize computation error between centralized and distributed estimation • Dynamic Link Assignment Problem (DLAP): - Find optimal communication topologies in real-time connecting local and inter-regional PDCs to maximize computational speed for the overall global estimation/monitoring/control problem. Swing Energy Function VE PDC a3s a3 s a3s a3 Weighted Consensus Distributed Subgradient Method ADMM Parallel variable method 2 Zero/First Order Hold Super PDC Fault-Tolerance & Cyber-Security Envisioned CPS architecture: 2 PMU PDC PMU PMU Swing Component of VPE PDC PDC PMU Tallahassee, FL PMU Kinetic Energy VKE PDC PMU PMU PMU PMU Clusters Proposed Distributed Cyber-Physical Architecture for PMU-PDC Communication: PDC Holyoke, MA Internet 1. Regional PMU data sent to regional PDC 2. Independent local analysis and storage 3. Regional PDCs send data to super-PDC for archival PDC Troy, NY Chicago, IL Control Actions Technical Approach State-of-art Centralized Processing Architecture: PMU Toronto, CA Princeton, NJ Cluster 1 PMU 0.22-Hz Mode Shape PMU Time Time Time Kinetic Energy Function Potential Energy Function P Ont ario P Michigan P New York • • P PJM • Total Swing Energy Software Console for Real-Time Digital Simulation Testbed • BEN-WAMS: Multi-vendor PMU-based hardware-in-loop simulation testbed at NCSU • High-fidelity dynamic models of IEEE 39-bus New England system & 115-bus WECC • PDC connected to 10 Gbps Breakable Experimental Network (BEN) • Execute distributed algorithms in BEN-WAMS • Test QoS, fault tolerance, cyber-security Distributed Voltage Monitoring Broader Impacts Construct PV curves from PMU data Voltage instability prediction over multiple areas Distributed estimation of voltage oscillations • Undergraduate, K-12 and minority education via Science House and FREEDM ERC programs at NC State • Undergraduate summer internship programs at Information Trust Institute at UIUC • Industry collaborations with power utilities and vendors such as SCE and ABB • Research Initiative Task Team outreach via NASPI