Ultra short term power prediction of offshore wind power based on ICEEMD-KPCA-LSTM

To improve the prediction accuracy of offshore wind power, an ultra-short-term wind power prediction method based on data decomposition and dimensionality reduction of input variables is proposed. First, the original wind speed, wind direction and temperature data are decomposed using the improved complete ensemble empirical mode decomposition (ICEEMD) to mitigate the data fluctuations and extract the internal hidden information. Secondly, in order to strengthen the correlation between the input variables and wind power and eliminate the redundant information, the extracted features were nonlinearly downscaled using Kernel Principal Component Analysis (KPCA) to retain the important feature information. Finally, rolling prediction of the reconstructed power components is carried out using long short term memory (LSTM), and the validity of the combination model proposed in this paper is verified by empirical analysis and selected comparison models. The experimental results show that the prediction method for offshore wind power based on ICEEMD-KPCA-LSTM is able to extract more key information within the wind power data, reduce the dimensionality of the input data, and effectively improve the prediction accuracy.