Research On Offshore Wind Power Generation Power Prediction Based On Machine Learning

Wind energy will become popular in the future as a significant low-carbon,sustainable,and clean energy source.Yet,because wind energy is unpredictable,it inevitably leads to offshore wind power’s instability,which has an impact on the power grid’s dependability and the reserve power plan.In order to provide technical support for the growth of the national electric power industry and grid maintenance personnel,it is necessary to accurately predict the power of offshore wind energy in advance.This paper’s research goal is to determine how to achieve accurate offshore wind power generation power prediction in advance,the primary research is described as follows:(1)A parallel combination forecasting model based on DAM,CNN-LSTM,and XGBoost is suggested in order to address the issue of unstable offshore wind power forecasting.First,the original offshore wind power data set is subjected to data quality control and feature correlation analysis;second,the model input data set is built using a time sliding window;third,an innovative attention mechanism is applied to the convolutional layer of CNN and the output layer of LSTM,that is,the dual-stage attention mechanism,which can produce noticeably better prediction results;and finally,following the concept of parallel prediction,use two LSTMs.The accuracy of this research model has been demonstrated through experiments to have a higher prediction value for offshore wind power;consequently,a set of software prediction system has been constructed using the suggested model as the research object.(2)This study suggests combining VMD decomposition and LSTM neural network model to predict offshore wind power generation power in order to address the instability of offshore wind power time series.The initial offshore wind power signal is split into its component parts and fused with LSTM and Light GBM,respectively.The final offshore wind power forecast result of the prediction model is obtained by merging the processed sub-components with the important features after screening for feature fusion,creating input variables using time sliding windows,and then predicting each sub-component.Experiments have shown that the prediction accuracy of the proposed fusion model has greatly increased,and that the prediction effect of offshore wind power signal after signal decomposition is significantly better than that without signal decomposition.