Learning-based Optimization of Active Distribution System Dispatch in Industrial Park Considering the Peak Operation Demand of Power Grid

The dynamic economic dispatch problem of the active distribution system combined of photovoltaic (PV), vanadium redox battery (VRB) energy storage device and multiple types of flexible load in industrial parks with uncertain renewable sources and demands is focused in this paper. First, the random dynamic variations of photovoltaic, multiple loads demand and peak operation demand are described as continuous Markov processes, and the VRB energy storage system is modeled considering its charge-discharge characteristics. Then, decision epoch, outputs level of photovoltaic, multiple load demands level, peak operation demands level and state of charge (SOC) level of VRB are defined as states of the system, the adjustment level of VRB and multiple types of flexible load are set as the actions. Based on relevant restrictions including the power balance constraint, the dynamic optimal dispatch problem for the system was described as a stochastic dynamic programming model, which aims to meet the peak operation demand of power grid and realize economic operation of the system. Finally, a reinforcement learning method is adopted to obtain the optimal policy. Simulation results show that the operational efficiency is significantly enhanced and the peak operation demand of power grid is partly satisfied by the optimal policy. Copyright © 2021 Acta Automatica Sinica. All rights reserved.