Cooperative grid frequency control under asymmetric V2G capacity via switched integral reinforcement learning

Abstract

In recent years, V2G techniques have been successfully implemented to stabilize frequency deviations in the grid caused by intermittent renewable energy sources. However, the battery capacity constraints and real-time dynamic driving demands of electric vehicles (EVs) in V2G systems hinder the coordination between V2G control and power plant frequency control strategies. It leads to the dynamic asymmetry of regulation up (RU) and regulation down (RD) frequency regulation (FR) capacity of V2G system, which can affect the performance of V2G FR. To cope with this problem, this paper proposes a switched integral reinforcement learning (SIRL) scheme for cooperative grid frequency control. The proposed scheme integrates V2G control with the power plant frequency control, which are cooperated with each other to complete FR tasks. A novel switched neural network (NN) structure is constructed for the training process of the SIRL algorithm to solve the problem of dynamic asymmetric FR capacity. A switched V2G utility function and a global cost function are constructed for RU and RD. Two independent subnetworks with the same structure are adopted for RU and RD learning to optimized V2G control parameters respectively under dynamic asymmetric output constraints. Two case studies of the IEEE 14-bus test system demonstrate the effectiveness and advantages of the proposed scheme.