Intra-day Load Forecasting Based On Multi-angle Feature Extraction And Combined Neural Network

Intraday load forecasting is a type of short-term load forecasting that plays an important role in power system planning and operation,and is an important basis for distribution system operation control,power sellers and virtual power plant operators to participate in intraday market transactions.With the transformation of China’s power system and the gradual opening up of the electricity market,the accuracy of intraday load forecasting is becoming increasingly important.However,under the new power system,the grid integration of flexible resources and the opening up of the power market will lead to a greater degree of uncertainty in the timing of load and a greater variety of influencing factors,making intraday load forecasting more difficult.This paper proposes an intraday forecasting method based on multi-angle feature extraction and combined neural network.For the intrinsic load feature extraction,the historical load data is first decomposed into sequences with different periods using SSA,and then the periods of each sequence are extracted using FFT.Finally,the intrinsic load feature set is constructed by capturing the previous cycle data in the original sequence at the time to be predicted and stitching it with the historical data.To extract the load extrinsic feature set,the correlation between the historical factors and the load is analysed using the enhanced GRA,and the highly correlated factors are finally filtered out to construct the load extrinsic feature set.Finally,the LSTM neural network and the multi-headed attention mechanism neural network were used to process the intrinsic load feature set and the extrinsic load influencing factor feature set respectively,and the results were fused by the fully connected neural network to obtain the load prediction values.Finally,the accuracy and stability of the proposed method were verified by ablation and control experiments to be better than traditional methods.