Short Term Power Load Forecasting Based on WSAE Deep Network Structure Model

Accurate and effective research on regional power load forecasting can not only help power supply enterprises develop reasonable power supply scheduling plans, but also reduce power losses in the system. Therefore, a highly accurate short-term power forecasting method is an important part of power system scheduling. This article takes a county in Xuzhou as an example to conduct research on short-term power load forecasting, qualitatively analyzing the impact of variables such as power load, weather, and time on load forecasting one by one. By specifying the input parameters for load forecasting and based on the stacked autoencoder (SAE) network framework, a deep network learning structure for wavelet stacked autoencoder (WSAE) is proposed. In order to compare the effectiveness of short-term power load forecasting, a comparative analysis was conducted with the power load forecasting models established by Extreme Learning Machine (ELM), Wavelet Neural Network (WNN), and Stacked Autoencoder (SAE). According to the evaluation data indicators, the performance of deep network models (WSAE and SAE) is superior to that of shallow network models (WNN and ELM). In the same deep network model, the WSAE model performs better than the SAE model, and the WNN model performs better than the ELM model.