考虑分布式光伏电站时空相关特性的集群功率预测OA

The rapid popularization of distributed photovoltaic(PV)power generation systems has made accurate power prediction particularly critical for stable grid operation and energy dispatch.However,compared with a single PV power plant,the prediction of distributed PV clusters is much more difficult because of the complex spatio-temporal correlations among the various sites,which are difficult to be effectively modeled by traditional methods.In order to improve the accuracy of cluster prediction,this paper proposes a method for distributed PV cluster power forecasting that incorporates representative station selection and Spatio-Temporal Graph Convolutional Networks(STGCN).Firstly,the spatio-temporal correlation among distributed PV stations is considered.After constructing the comprehensive correlation coefficients,the output power of each distributed PV station is taken as a distinct feature affecting the cluster power.Representative stations are selected based on the Maximum Relevance Minimum Redundancy(mRMR)feature selection method.Then the Hiking Optimization Algorithm(HOA)is introduced to portray the contribution of the output power from each representative station to the cluster power,enabling the calculation of optimal weight allocation for each representative power station.Finally,a spatio-temporal graph convolutional network is constructed using the historical data from the representative stations to realize the feature extraction of complex spatio-temporal data among PV stations.After outputting the power prediction data of each representative station,the cluster prediction power is obtained through optimal weight calculations.Example calculations are conducted in two PV clusters in S1 and S2 provinces to prove the effectiveness of the proposed method in cluster prediction.