Multi-device cooperative reactive power optimization control strategy for high percentage distributed photovoltaic distribution grid based on proximal strategy optimization algorithm

A high proportion of strongly intermittent distributed PV is connected to the distribution network in a decentralized and disordered manner, which leads to the difficulty of its reactive power balance. This paper proposes a multi-device collaborative reactive power optimization control method based on proximal strategy optimization, which models the voltage reactive power optimization problem of high-ratio distributed PV distribution grids as a partially observable Markov decision-making process, and then establishes a method of generating and extracting training samples under the framework of deep reinforcement learning, and uses proximal strategy. The algorithm dynamically controls the update amplitude of the network through the shear strategy gradient method and improves the sample utilization through importance sampling, which has good convergence when dealing with continuous action space and high-dimensional state space. Finally, arithmetic simulations and analysis are performed based on the IEEE 33-bus system. The results show that the algorithm learns a policy network that can solve the global optimal solution based on the electrical information of the respective nodes.