Wind power error compensation prediction model based on CEEMD-SE-ELM-TCN

Abstract

Abstract Wind power generation holds immense importance in addressing the issue of global energy shortage, while precise wind power forecasting proves essential for effective management and dependable operation of wind power networks. This study introduces a hybrid deep learning model, encompassing complete ensemble empirical mode decomposition (CEEMD), sample entropy (SE), extreme learning machine (ELM) and time convolutional network (TCN), for accurately predicting short-term wind power output. First, CEEMD decomposed the original wind power into multiple submodes, which effectively reduced the series volatility. Then, the SE of intrinsic mode function sequence is calculated, and the subsequences with similar complexity are superimposed to reduce the calculation cost, improve the simulation accuracy and reduce the noise of the original wind power sequence. Secondly, the ELM model is established for each submode, and the prediction error of BiLSTM is predicted again using TCN to improve the efficiency and prediction performance of the hybrid model. Finally, the outcomes of each individual submode are amalgamated to yield the ultimate prediction outcome. To showcase the efficacy and dominance of the error compensation technique, several comparison models were established in the experiment. The results demonstrated that the suggested hybrid model exhibits superior predictive accuracy in the domain of wind power prediction. Compared with the comparison model, the improvement in MAPE and RMSE was 60.50 and 77.74%, respectively.