Research On Short-term Wind Speed Prediction Method Based On Deep Learning

In recent years,with the rapid development of my country’s industrialization,informatization,urbanization and agricultural modernization,the scale of the power system has continued to expand,and the demand for finding and utilizing clean,safe,economical,and renewable energy has become increasingly urgent.Wind energy is an important clean,safe,economical,and renewable energy source,and wind speed is one of the biggest factors affecting wind power generation systems.However,due to the strong randomness of wind speed,it is difficult to integrate wind power into the grid.Improving the accuracy of wind speed prediction is conducive to the optimal scheduling of wind power generation systems,and obtaining reliable prediction uncertainty information is also conducive to avoiding planning risks.In general,it is very important for the application of wind energy to obtain wind speed prediction results with high accuracy,high reliability and suitable prediction range.This study is committed to promoting and solving the current research difficulty of wind energy utilization,namely,how to achieve effective prediction of wind speed.The main work includes three aspects: constructing a deep learning model represented by the space time fusion mechanism of convolutional and shared weight coefficient long and short time memory networks for short-term wind speed point prediction,and studying an improved deep learning method represented by the space time fusion mechanism of convolutional and shared weight coefficient long and short time memory networks and GPR for short-term wind speed interval prediction,the improved deep learning prediction method proposed in this study has performance advantages over existing methods,verified through the prediction of measured wind speed datasets..This study aims to fully extract the spatial and temporal features of wind speed data for short-term wind speed point prediction.A spatiotemporal fusion mechanism based on convolutional and shared weight coefficient short-term memory network,CSWLSTM,is constructed to solve the problem of difficulty in improving the accuracy of short-term wind speed point prediction.Experimental results have shown that the improved deep learning wind speed point prediction model proposed in this study can achieve end-to-end adaptive data processing,resulting in predicted values that are closer to observed values and higher prediction accuracy.It has significant advantages over existing methods in wind speed point prediction.This study focuses on complex scenarios that need to reflect the uncertainty components in the prediction results.Using the high-precision point prediction results obtained from the improved deep learning point prediction model constructed in this study as input,combined with the machine learning prediction method GPR,the improved deep learning interval prediction method CSWLSTM-GPR is proposed for short-term wind speed interval prediction and uncertainty evaluation,Solve the problem of traditional machine learning prediction methods being unable to accurately describe the volatility and uncertainty of wind speed data through data-driven methods.Experiments have shown that the improved deep learning wind speed interval prediction method proposed in this study can obtain more suitable prediction intervals,higher prediction reliability,and more obvious performance advantages.This study constructs a short-term wind speed prediction model based on improved deep learning,and implements experimental verification of actual wind speed data prediction in a certain wind farm on Python.The improved method point prediction result indicators and interval prediction result indicators are obtained.Experiments have shown that the improved deep learning prediction method CSWLSTM-GPR proposed in this study has the best prediction performance.It not only automatically extracts and selects features,but also has stable prediction performance and good network performance,solving the problems of low prediction accuracy,weak interval prediction,and robustness of traditional methods.