Dynamic Optimization of on‐Grid Integrated Energy System Considering Peak‐Shaving Demand Via Learning Methods

Abstract

An integrated energy system (IES) is combined with electricity, gas and heating subnetworks to satisfy various loads, which results in a complicated energy coupling. Considering the power interaction between IES and power grid, the real-time dynamic optimal dispatch policies for the on-grid IES are proposed and compared by learning methods to improve the operation economic and peak-shaving performances in this study. Owing to the uncertainties of the load demands and real-time peak-shaving demand and renewable energy, together with the sequential operation of the controllable sources, the energy dispatch optimization problem of the IES is described as a stochastic dynamic optimization process. A source-load collaborative operation mode is proposed as the coupling of thermal energy and electrical power in the IES and compared with different energy dispatching mode. And reinforcement learning methods are adopted to achieve the dynamic optimal policy. Since the multiple systems with energy coupling would result in modeling difficulty and dimension curse during the optimization, a deep reinforcement learning method (DDQN) is used to solve this problem. A numerical analysis is performed by comprehensively comparing the different learning algorithms and operation modes, and the simulation results show that the derived dispatch policy achieved by DDQN method with source-load collaboration can effectively improve the performance of the IES in the stochastic environment. © 2022 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC.