Deep Reinforcement Learning Method For Operation Optimization Of Active Distribution Network

With the increasing penetration of distributed generation(DG),flexible load and other new elements in the distribution network and the construction of distribution automation system and information system,the traditional distribution network is gradually evolving into an observable and controllable active distribution network(ADN).At the same time,the uncertainty of system and environment brings new challenges to the operation of ADN.Deep reinforcement learning(DRL)can obtain control strategies directly from real-time environment operation information by replacing process simulation with data-driven.This paper studies the application of DRL method in ADN operation optimization,which is of important significance in theory and practice.Firstly,the overall framework of ADN operation optimization based on the dueling deep Q network(DDQN)is constructed,and then the DRL method is used for self-learning optimization aiming at the demand response(DR)management of interruptible load(IL)and the fault recovery control of ADN under extreme disaster.Finally,the obtained strategy is verified by simulation examples.The main innovative achievements of the paper include:1.The overall framework of ADN operation optimization based on DDQN is established,in which the ADN operation is modeled as the Markov decision process.Firstly,the observation state,control action and immediate reward functions are defined respectively according to specific control objectives and constraints.Then,the DRL algorithm is applied to train the DDQN network to approximate the estimation function.Finally,the operation optimization strategy of ADN can be generated automatically according to the real-time observation state.2.In order to ensure the economical operation of ADN,the DDQN-based automatic demand response architecture of IL is proposed,and the DRL optimization is carried out to realize the direct mapping from the real-time operation state to the DR strategy with the goal of regulating voltage and reducing operation costs.The obtained DR strategy can reduce both the peak load demand and the operation costs on the premise of regulating voltage to the safe limit under the time of use tariff and variable electricity consumption patterns.3.In order to enhance the resilience of ADN under extreme disasters,the fault recovery control framework of ADN based on DDQN is constructed,and the DRL algorithm is applied to realize the direct mapping from the real-time environment and operation state to the fault recovery strategy with the goal of enhancing resilience and improving operational economy,which can restore the power supply of load to the maximum extent in all states under uncertain external environment(such as fault location)and system operation(such as intermittent DG and fluctuating load).