Short-Term Power Load Forecasting Based on CNN-LSTM-Attention Modeling

Accurate short-term power load forecasting is crucial for the safe, stable and economically efficient operation of power systems. Aiming at the limitations of traditional prediction methods in feature extraction and data mining, which are difficult to meet the demand for high-precision load forecasting, a short-term power load forecasting model integrating convolutional neural network (CNN) and long short-term memory (LSTM) with attention mechanism is proposed. Among them, the CNN module is used to extract the deep features in the load sequence, the LSTM module is responsible for capturing the temporal dependencies of loads, and the attention mechanism highlights the influence of key factors by dynamically allocating the weights of input features. Based on the simulation analysis of year-round power load data in a region in southern China, the results show that compared with the traditional forecasting methods, the CNN-LSTM-Attention model can more accurately capture the seasonal changes of power loads, improve the forecasting accuracy and computational efficiency, and show strong adaptability and stability in different seasons, so as to realize more accurate short-term power load forecasting, which has a high engineering application value.