MPC and Forecasting for Smart Energy Systems

Transcription

1 MPC and Forecasting for Smart Energy Systems John Bagterp Jørgensen Leo Emil Sokoler, Laura Standardi, Rasmus Halvgaard, Tobias Gybel Hovgaard, Peter Juhler, Dinesen, Gianluca Frison, Caroline Johannsen CITIES Workshop WP 5 Forecasting and Control Grundfos December 9, 2014

2 Smart Energy Systems Thermal Storage Heating of floors etc Heating of water accumulation tanks Refrigeration Systems 2 DTU Compute, Technical University of Denmark Power / Heat Producers Wind Turbines Photovoltaic Solar Modules Solar Panels CHP Plants Fuel Cells

3 Connected Energy Systems 3 DTU Compute, Technical University of Denmark

4 Electricity & Heating / Cooling 4 DTU Compute, Technical University of Denmark

5 Multiple Connected Energy Hubs 5 DTU Compute, Technical University of Denmark

6 Model Predictive Control 6 DTU Compute, Technical University of Denmark 6

7 Moving Horizon Control Past Predicted Future Setpoint Predicted Output Input Time 7 DTU Compute, Technical University of Denmark

8 Moving Horizon Control Past Predicted Future Setpoint Predicted Output Input Time 8 DTU Compute, Technical University of Denmark

9 Moving Horizon Control Past Predicted Future Setpoint Predicted Output Input Time 9 DTU Compute, Technical University of Denmark

10 Moving Horizon Control Past Predicted Future Setpoint Predicted Output Input Time 10 DTU Compute, Technical University of Denmark

11 Moving Horizon Control Past Predicted Future Setpoint Predicted Output Input Time 11 DTU Compute, Technical University of Denmark

12 Moving Horizon Control Past Predicted Future Setpoint Predicted Output Input Time 12 DTU Compute, Technical University of Denmark

13 Moving Horizon Control Past Predicted Future Setpoint Predicted Output Input Time 13 DTU Compute, Technical University of Denmark

14 Optimal Operation is Close to Limits P2 Optimum operation point Limits Optimum close to constraints requires process optimization and advanced process control P1 14 DTU Compute, Technical University of Denmark

15 Economic Value of Process Optimization Limit Safety Margin Target No APC APC reduces variation Reduced variation allows operation closer to the limit DTU Compute, Technical University of Denmark

16 Economic Value of Proces Optimization Economic value added by Process Optimization DTU Compute, Technical University of Denmark Squeze & Shift

17 Manipulated Variable Controlled Variable Rapid Product Change Time 18 DTU Compute, Technical University of Denmark 1

18 Manipulated Input Measured Output Step Response Experiments DTU Compute, Technical University of Denmark

19 Manipulated Input Measured Output Step Response Experiments K( s 1) s Gs () e ( T s 1)( T s 1) DTU Compute, Technical University of Denmark

20 Identification of the 21 Deterministic and Stochastic Model 21 DTU Compute, Technical University of Denmark

21 Drivers of MPC Powerful Computers Mathematical / Statistical Modelling Optimization Algorithms 22 DTU Compute, Technical University of Denmark

22 Economic MPC Mathematical Optimization 23 DTU Compute, Technical University of Denmark

23 Economic MPC for Uncertain Systems 24 DTU Compute, Technical University of Denmark

24 Bi-Criterion Economic MPC 25 DTU Compute, Technical University of Denmark

25 Risk Mitigation (Regularization) 26 DTU Compute, Technical University of Denmark

26 Hierarchical Control Structure 27 DTU Compute, Technical University of Denmark

27 Fast Solver for Direct Control of an Entire City 28 DTU Compute, Technical University of Denmark

28 Control of Smart Energy Systems = Economic MPC 29 DTU Compute, Technical University of Denmark

29 Virtual Power Plants Controlling the Power Demand 30 DTU Compute, Technical University of Denmark

30 31 DTU Compute, Technical University of Denmark Faroe Islands

31 Frequency Dynamics at a Contingency Event 32 DTU Compute, Technical University of Denmark

32 The Idea of Coupling the Unit Commitment and Economic MPC Day-ahead planning: Prepare power production schedule the next 24 hours based on available forecast on tomorrow demand load and power production from renewable sources Online Control: Online framework adapts during the day the predefined schedule based on new and more reliable forecast on power production from renewable sources The key ideas as we will see is that the Unit Commitment problem should be solved as often as possible or coupled with the Economic MPC. To this requires better numerical methods than available today 33 DTU Compute, Technical University of Denmark

33 Unit Commitment & Economic MPC 34 DTU Compute, Technical University of Denmark

34 Unit Commitment & Economic MPC 35 DTU Compute, Technical University of Denmark

35 Supermarket Refrigeration Control variables Disturbances 36 DTU Compute, Technical University of Denmark

36 Supermarket Refrigeration 37 DTU Compute, Technical University of Denmark

37 38 DTU Compute, Technical University of Denmark Case Study

38 39 DTU Compute, Technical University of Denmark House with Heat Pump

39 Temperature Control Danfoss NEST (Google) 40 DTU Compute, Technical University of Denmark

40 Smart Grid Ready Heat Pump 41 DTU Compute, Technical University of Denmark 41

41 Schematics of House with Heat Pump 42 DTU Compute, Technical University of Denmark

42 Economic MPC for Building Control 43 DTU Compute, Technical University of Denmark

43 MPC & Process Optimization Toolbox Features Simplicity - easy to Commission Tune Maintain Customizable and adaptable to Process dynamics Process modifications Operational strategies Includes frontier technologies in Mathematical Optimization Process Control Software Engineering Mathematical/Statistical Modeling and Simulation 44 DTU Compute, Technical University of Denmark

44 The Extended LQ Problem 45 DTU Compute, Technical University of Denmark

45 KKT System for the Extended LQ Problem 46 DTU Compute, Technical University of Denmark

46 KKT System for the Extended LQ Problem 47 DTU Compute, Technical University of Denmark

47 Software Implementation Rasperry PI C/C++ Smart Phone C/Java/Matlab 48 DTU Compute, Technical University of Denmark

48 Project Plan WP 5 CITIES Hire Post Doc Call and hire Hire PhD Identify candidates Do MSc Project Post Doc 24 Months WP 5.1 Stochastic Simulation Models Stochastic Linear and Nonlinear Models Wind Turbines Vestas Photovoltaics??? CHP Plants DONG Buildings DTU Civil Eng Refrigeration Systems Danfoss HVAC Systems Danfoss Electrical Vehices??? Fuel Cells Topsøe Fuels Cells Electrolysis Risø Accumulation Tanks DONG Micro Power Plants??? Linear model spec/sim: TFs, SSs, Stochastic, Delay Nonlinear model spec/sim. ODE,SDE Power Hub concept WP 5.2 Forecasting Methods Stochastic Differential Equations Filtering and prediction using SDEs - optmization Covariance, probability distribution, Filtering and prediction using SDEs - ensemble Experimental Design Methods Monitoring and Fault Detection Test for Different Energy System Components PhD 36 Months Control using probabilistic input WP 5.3 Control Methods using Probabilistic Info Course work Certainty equivalence MPC with back-off Teaching assistant Probabilistic control of linear systems Litterature study Mean-variance control of linear/nonlinear systems Programming and comparing prototypes Condition-VaR control of linear/nonlinear systems Define a set of test cases Stochastic Programming MPC Tuning methods Test of these methods for energy components WP 5.4 Indirect Control Methods Distribution based methods Optimization based methods Test of these methods for energy components Indirect Control Methods PhD / Post-Doc / Faculty / External?? WP 5.5 IT / Portal for forecast and control Archictecture Programming Test IT Tools 49 DTU Compute, Technical University of Denmark

49 Thank You Q & A John Bagterp Jørgensen Technical University of Denmark jbjo@dtu.dk 50 DTU Compute, Technical University of Denmark