Ultra-short Term Wind Power Prediction method based on MOGWO-VMD-LSTM Algorithm

Wind power has the characteristics of strong volatility and randomness. In order to improve the prediction accuracy, a combined wind power prediction method based on multi-objective grey wolf algorithm (MOGWO), variational mode decomposition (VMD) and long short-term memory (LSTM) was proposed. Firstly, in order to consider the subsequence that balances the decomposition accuracy and feature validity by using subsequence permutation entropy and root mean square error were used as the fitness functions of MOGWO, and obtaining the VMD parameters were optimized by MOGWO. Then, the respective LSTM models were built for the decomposed subsequences, and finally the subsequence predictions were summed to obtain the final results. Field data from a wind farm was selected for test, then the wind power for the next four hours was predicted to assess the point at the fourth hour. Compared with the LSTM and other displayed algorithms, the results show that the MOGWO-VMD-LSTM based method has better prediction performance, and its accuracy can achieve up to 86.87%.