A Data-Driven Coordinated Active and Reactive Dispatching Strategy for Photovoltaics

The influx of photovoltaic systems brings about fluctuations on the grid and risks of overvoltage, issues which may be remedied by fully exploiting the reactive power support capabilities of inverters. However, reliable inverter operation is closely tied to the condition of its Insulated Gate Bipolar Transistor (IGBT) junction temperatures. In this paper, we propose an active-reactive coordinated control strategy called AGVC that considers the dependability of inverter components to simultaneously enhance the response of AGC and AVC. Firstly, we introduce a data-driven method for assessing component dependability which employs a temporal convolution network (TCN) model to calculate IGBT junction temperature data. Then, we establish an active-reactive coordinated optimization model for the distribution network with attention to component dependability, introducing the value of IGBT junction temperatures into the optimization objective. Finally, we complete agent training using a gradient reinforcement learning algorithm of multi-agent deep deterministic policy. The IEEE 33-node system demonstrates the advantages of the proposed strategy in improving active dispatch and voltage balance compared to recent research methods, with average active regulation error and voltage stability enhanced.