Development approach of a volt-var control for inverter-coupled renewable energy plants using deep reinforcement learning

The connection of RES and the disconnection of controllable synchronous generators pose new challenges for grid management and steady-state voltage control. Removed reactive power sources that have contributed to voltage stability need to be replaced. One approach is the use of inverter-coupled RES as reactive power sources and the application of Artificial Neural Networks (ANN) for optimised reactive power provision. Several studies have already started to explore the possibilities of Deep Reinforcement Learning (DRL) to develop new reactive power control methods for RES. While other studies have mostly looked at the control of a single unit in a low-voltage grid, this study develops a volt-var control for multiple RES investigated in different operating scenarios for a whole year in a CIGRE benchmark medium-voltage distribution grid. The results show that the DRL-trained ANN can control the reactive power supply with a lower voltage deviation compared to conventional methods. However, this is at the cost of more frequent equipment overloads due to an increase in reactive power flows. The application of this developed approach will be further explored in order to validate the methodology in real grid models in the future.