A Random Forest-Assisted Fast Distributed Auction-Based Algorithm for Hierarchical Coordinated Power Control in a Large-Scale PV Power Plant

Abstract

In response to the automatic generation control (AGC) signals, the direct power control of a large-scale PV power plant easily encounters the communication bottlenecks, the high optimization difficulty and computation burden due to the large number of controllable inverters with different response performances. To handle these problems, a hierarchical framework of coordinated power control (CPC) is constructed, which is decomposed into a upper-layer CPC between different sub-areas and a lower-layer CPC between different inverters in each sub-area. Instead of a centralized optimization, a novel random forest-assisted fast distributed auction-based algorithm (FDAA) is proposed for a distributed optimization of CPC. The random forest can rapidly generate a dynamic surrogate model of the optimization results from the low-layer CPC to the upper-layer CPC, thus these two-layer optimizations of CPC can be decoupled without too much interactions and computations. The effectiveness of the proposed method is thoroughly evaluated on a PV power plant with 10 sub-areas and 100 inverters under various irradiation conditions.