Files in this item



application/pdfUILU-ENG-12-2204_DC-256.pdf (409kB)
(no description provided)PDF


Title:Coordinating Dispatch of Distributed Energy Resources with Model Predictive Control and Q-Learning
Author(s):Kowli, Anupama; Mayhorn, Ebony; Kalsi, Karanjit; Meyn, Sean P.
Subject(s):Approximate dynamic programming
Distributed energy resources
Dynamic dispatch
Energy storage
Model predictive control
Power grid
Reinforcement learning
Abstract:Distributed energy resources such as renewable generators (wind, solar), energy storage, and demand response can be used to complement fossil-fueled generators. The uncertainty and variability due to high penetration of renewable resources make power system operations and controls challenging. This work addresses the coordinated operation of these distributed resources to meet economic, reliability, and environmental objectives. Recent research proposes Model Predictive Control (MPC) to solve the problem. However, MPC may yield a poor performance if the terminal penalty function is not chosen correctly. In this work, a parameterized Q-learning algorithm is devised to approximate the optimal terminal penalty function. This approximate penalty function is then used in MPC, thus effectively combining the two techniques. It is argued that this combination approach would lead to the best solution in terms of computation, and adaptability to a changing environment. Simulation studies demonstrating the efficacy of the proposed methodology for power system dispatch problems are presented.
Issue Date:2012-05
Publisher:Coordinated Science Laboratory, University of Illinois at Urbana-Champaign
Series/Report:Coordinated Science Laboratory Report no. UILU-ENG-12-2204, DC-256
Genre:Technical Report
Sponsor:National Science Foundation / CPS-0931416
Department of Energy / DE-OE0000097 and DE-SC0003879
Pacific Northwest National Laboratory
Date Available in IDEALS:2016-07-06

This item appears in the following Collection(s)

Item Statistics