Learning Based Coordinated Voltage Control in PV Rich Distribution Systems

As the proportion of distributed photovoltaic (PV) power generation in the distribution network gradually increases, the resulting voltage regulation problem becomes increasingly complex. In this paper, a voltage control method based on Recurrent Deep Reinforcement Learning (RDRL) is proposed to achieve the cooperative operation of different voltage management devices. Starting from the action logic and control rules of On-Load Tap Changer (OLTC) and the principle of voltage regulation by PV inverters through reactive power compensation, this paper analyses the cooperative operation mechanism of these two devices in the distribution network and points out the problems that may occur when there is a lack of cooperation. To address this problem, this paper uses the Recurrent Markov Decision Process (RMDP) algorithm to achieve synergistic voltage control of OLTC and PV inverters by controlling the reactive power output of PV inverter clusters. In the technical method section, this paper describes the RMDP model considering entropy and transforms the voltage control problem into an RMDP problem. To demonstrate the effectiveness of the method proposed in this paper, the proposed RMDP algorithm is simulated and tested under strong photovoltaic fluctuations using the IEEE-33 node distribution network model. The test results show that the RMDP algorithm can effectively collaborate with the OLTC and PV inverters for voltage control, reduce the number of OLTC tap actions, reduce loss of the distribution network, and keep the voltage within the permissible range.