Load Forecasting Method Based on SSA-DAP-TCN

As the sampling frequency of electricity load increases, noise is difficult to filter at high sampling rates, which can lead to deviation in load forecasting and affect the performance of prediction algorithms. Enhancing the accuracy of load forecasting is crucial to formulate production plans, guide economic dispatch and maintain the balance of power supply and demand. This study proposes a power load forecasting method based on multiscale decomposition and reconstruction. Initially, the load data is decomposed into trend, periodic, and noise components by singular spectrum analysis (SSA), effectively eliminating noise. Then, the affinity propagation algorithm based on dynamic time warping distance (DAP) is utilized to reconstruct the components to obtain new subsequences. Time convolutional network (TCN) is utilized to predict the reconstructed subsequences, and finally the forecasted results of each subsequence are accumulated to get the final forecasted results. To verify the superiority of the proposed method, this study take actual manufacturing industry load data as an example and comparing it with various prediction models. Compared with TCN, the root mean squared error of the proposed method is improved by 55.68%.