Technische Universität München Distributed Optimization for Smart Grids Jose Rivera , Christoph Goebel, and Hans-Arno Jacobsen Smart Buildings and Smart Grids Dagstuhl Seminar, February 22 – 27 , 2015 Department of Computer Science Chair for Application and Middleware Systems (I13)
Technische Universität München Scenario • Very large number of (new) devices • Smart grid allows control Can we actively integrate them into power system operations in a scalable , efficient and reliable manner ? This usually involves solving optimization problems [ frauenhofer-esk ] [ionexusa.com] [energiewende-sta.de] [smartgridsmartcity_com_au] 2 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Centralized optimization Computation Data 1. All data to one solver 2. Solver optimizes 3. Results are send back Control center 4. Done! (Aggregator) Challenging to scale Devices up to large number of devices ! 3 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Hierarchical optimization Idea: Simplify problem by splitting or creating a Computation hierarchy of problems 1. Each solver gets part Data of the data 2. Each solver optimizes 3. Each solver sends Control center result back (Aggregator) 4. Done! Scales better than centralized, but offers Devices suboptimal results 4 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Distributed optimization Idea: Divide into subproblems and one coordinator Computation 1. Each subproblem gets Data local data Aggregator 2. Coordinator sends coordination signal coordinator 3. Each solver optimizes 4. Each solver sends result to coordinator Devices 5. Coordinator updates subproblems coordination signal 6. If no convergence go to 2 7. Done! Scalable and optimal ! 5 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München An example: EV ADMM • General algorithmic framework for aggregator convex optimization (use case: EV charging) ADMM General aggregator optimization problem 6 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Using distributed optimization for the smart grid • Some issues – Efficiency: Can have expensive network and cpu usage* – Reliability: No results until convergence (important for communication delays) • Needed – Killer app – Platform for simple formulation and deployment *MapReduce/Bigtable for Distributed Optimization: http://videolectures.net/nipsworkshops2010_guestrin_kml/ 7 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München DOPS: Distributed Optimization Pub/Sub Publish/Subscribe middleware for www.msrg.org/padres/ EV aggregation • Performance improvement based on DOs characteristics • Communication with coordinator is a bottleneck • In-network computation: Idea: In-network 2 x faster distributed optimization! computation 8 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Reliability: EV charging control Problem: Real-time capability of distributed optimization Idea: Anytime algorithms Result: EV charging control can run in milliseconds instead of minutes Distributed optimization feasible on each iteration 9 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Overview: Distributed optimization for smart grids • High potential for the definition of control protocols and practicable system-wide optimization • Currently it may introduce more problems than it solves • Becomes interesting when centralized approach starts to be intractable • Suitable for planning tasks, but close-loop control remains a challenge • A lot of theory but very limited actual implementation 10 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Crowdsourcing grid data Collection Verification Inference Analysis Transformer? Generator? EnergyMap.info Expert in the loop 11 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München References (1/2) C. Goebel et al., “Energy Informatics,” Business & Information Systems Engineering, 2013. Matt Kraning, Eric Chu, Javad Lavaei and Stephen Boyd , " Dynamic Network Energy Management via Proximal Message Passing ", Foundations and Trends in Optimization: Vol. 1: No. 2, pp 73-126. Boyd, Stephen, et al. "Distributed optimization and statistical learning via the alternating direction method of multipliers." Foundations and Trends in Machine Learning 3.1 (2011): 1-122. L. Gan, U. Topcu, and S. Low. Optimal decentralized protocol for electric vehicle charging . IEEE Transactions on Power Systems, 2013. Omid Ardakanian, Catherine Rosenberg, and S. Keshav. 2013. Distributed control of electric vehicle charging . In Proceedings of the fourth international conference on Future energy systems (e-Energy '13). ACM, New York, NY, USA, 101-112. 12 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München References (2/2) J. Rivera, M. Jergler, A. Stoimenov, C. Goebel, H.-A. Jacobsen. Using Publish/Subscribe Middleware for Distributed EV Charging Optimization. Conference on Energy Informatics, Zürich, Switzerland, 2014 . J. Rivera, H.-A. Jacobsen. A Distributed Anytime Algorithm for Network Utility Maximization with Application to Real-time EV Charging Control. 53rd Conference on Decision and Control (CDC2014), Los Angeles, USA. 2014. J. Rivera, P. Wolfrum, S. Hirche, C. Goebel, and H.-A. Jacobsen. " Alternating direction method of multipliers for decentralized electric vehicle charging control ." In 52nd Conference on Decision and Control (CDC2013), Florence, Italy, 2013 . 13 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Thank you, questions or comments … Contact Jose Rivera j.rivera@tum.de 14 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Outline: Power system operations What is the status quo? Why do we need to do things differently? Our contribution Searching for a killer-app Summary and future work 15 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Power system operations are complex [gettyimages] 16 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Power system operations Operation Decision Monitoring Analysis enhancement support • Status & Analog • Dispatcher Power • Optimal Power Flow • Interlocking with LF & Retrieval (SAR) Flow (DPF) (OPF) SA • Network Model Builder • Security Analysis (SA) • Security Constrained • Study data base (NMB) • Short Circuit Analysis Dispatch (SCD) • Network save cases • Scheduler Function • Voltage Stability (SCA ) (SF) Analysis (VSA) • State Estimation (SE) • Thermal Security • Network Sensitivity Analysis (TSA) • Available (NS ) Transmission Capacity (ATC=VSA+TSA) • Equipment Outage Scheduler (EOS) • Bad Topology Detection (BTD) • Network Parameter Update (NPU) • Network Modeling Assistant (NMA) [G. Björkman, ABB] 17 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Power system operations & time scales Real-Time Operations (msec – 10s of minutes) • Protection (msec) • Frequency governors (sec) • Automatic Generation Control (AGC) (seconds) • State estimation and contingency analysis (minutes) • Economic dispatch (~15 minutes) Operation Decision Monitoring Analysis enhancement support Operation planning (days - years) • Load forecasting – days (short term) to years (long term) • Unit commitment (day ahead markets) • Maintenance planning (weeks - year) • Generation and transmission planning ( up to 25 years) 18 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München In simple words Power system operations = Optimization 19 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Conventional is Methods preferred Neural Nonlinear Programming Network (NLP) (NN) Linear Interior Point Programming (IP) methods (LP) Intelligence Conventional Particle Mixed - Evolutionary Quadratic search Swarm Integer optimization Algorithms Programming Programming Optimization (QP) (EAs) (MIP) (PSO) methods methods Generalized Network Flow reduced Programming gradient (NFP) method Newton Tabu Search method (TS) 20 Department of Computer Science, Chair for Application & Middleware Systems
Technische Universität München Why not heuristics? • No guaranteed good results • Also… 21 Department of Computer Science, Chair for Application & Middleware Systems
Recommend
More recommend
