Electric Distribution Network With Multi-Microgrids Management Using Surrogate-Assisted Deep Reinforcement Learning Optimization

Abstract

Increasing of electric vehicle demand and uncertain electricity generation from solar photovoltaics lead to poor reliability in the Distribution Network (DN). MicroGrids (MG) connecting to the DN with suitable energy trading is an effective way to solve this issue. However, suitable energy trading considering grid constraints with a low computational burden is a challenge for solving Optimal Energy Management (OEM). Therefore, this paper proposes an OEM using a surrogate which is modeled based on a Deep Neural Network to assist the Deep Reinforcement Learning Optimization for reducing the computational burden. The proposed method is applied to a bi-level OEM for multi-MGs connected in the DN with real-time pricing consideration, represented as the proposed strategy. The surrogate is modeled to predict the power system parameters of the MG based on probabilistic power flow, whereas a deterministic power flow is applied to evaluate the parameters of DN. To validate and demonstrate the proposed method and strategy, the modified IEEE-33 test feeder and a residential distribution system which are defined as the DN and MG, respectively are used to test. Simulation results show that the proposed method can reduce computational burden by 89.23% compared with the Differential Evolution algorithm. Moreover, the proposed strategy provides the optimal purchased energy price of DN offering to each MG which can reduce the cost of energy purchased from the main grid resulting in lower operating cost of DN.