An optimized deep nonlinear integrated framework for wind speed forecasting and uncertainty analysis

Accurate wind speed prediction improves the efficiency of electricity and also increases the economic benefits of wind farms. However, the previous studies do not fully consider the chronological feature of wind speed, and do not properly identify and use the characteristic decomposed components of wind speed series in different time periods. The phenomenon of insufficient and excessive decomposition of components weakens the prediction accuracy of the model and destroys the power generation plan of the wind farm. In this paper, a segmented multi-modal deep learning integrated model based on periodicity is proposed. Firstly, the time series is reconstructed into matrix according to date; the matrix is divided into different segment matrices according to chronological characteristics; each sub-matrix is spliced into sub-sequence according to chronological order. Secondly, the contribution extremum method is used to determine the optimal feature components of all sub-sequences. Thirdly, the optimized deep learning model integrates all sub-sequences and then obtains the prediction results of the current date. Finally, the uncertainty analysis of the predicted value further improves the reliability of wind speed prediction. The data of wind farm in Hexi corridor area of China are used to simulate the experiment, and the results show that the proposed model has good performance. •A time series segmentation method based on periodicity is proposed.•The contribution extremum is used to optimize the component decomposition method.•Optimized deep learning prediction model to achieve feature integration.•Uncertainty analysis enhances the persuasion of point prediction model.