Multi-channel short-term electric load forecasting model based on CEEMD data decomposition

Aiming at the stochasticity and instability of power load data and the difficulty of traditional forecasting models to accurately capture the local and global features of the data, a multi-channel short-term power negative character model (CACCLA) based on CEEMD is proposed. Firstly, the method utilizes CEEMD to decompose the original load teachings into feature modal components representing the unused scales to reduce the instability of the original series; then, multi-features are introduced into each modal component after the decomposition, and the Bagging builds multiple models through self-sampling, it effectively reduces the variance of predictions and improves stability and accuracy.; and finally. The CNN-CBAM+LSTM-ATT model was used for load prediction. Through comparison experiments with several prediction models, the results show that the model has a better prediction effect than traditional prediction methods.