Research On Multi-model Wind Speed Prediction Based On Variable Selection Of Wind Field Big Data

The vast land and various landforms endow China with abundant wind energy resources.In recent years,with the adjustment of the national energy strategy,wind power generation has a broad prospect.Accurate and fast wind speed prediction is of great significance to improve the quality of wind power and ensure the safe and stable operation of power grid.The current wind speed forecasting field needs to be improved in terms of data screening,variable selection strategies and the accuracy of wind speed forecasting models.In order to achieve multi-sensor data fusion and further improve the accuracy of wind speed prediction models on the basic of existing models,this paper conducts multi-model wind speed prediction research based on the SCADA historical data of a wind field in North China.The content is as follows:In order to solve the problem of outliers in the historical data of wind field SCADA,the quartile method wind field big data identification strategy based on wind speed and wind power characteristic curve is adopted.Through the use of two times of the quartile method,the outliers hidden in the wind field big data are cleaned effectively,and the data quality is improved.Considering the lengthy data of wind field SCADA,a combined variable selection strategy based on random forest algorithm and maximum information coefficient algorithm is proposed to analyze the correlation of more than 40 variables.This method effectively improves the reliability of variable screening results,reduces the input variable dimension of the model,and lays a data foundation for establishing the wind speed prediction model.To improve the prediction accuracy of single artificial neural network model,a BP neural network(BAS-BP)wind speed prediction model optimized by beetle antennae algorithm is established.By simulating three discontinuous test samples in the time dimension,and comparing with a single BP model,genetic algorithm optimized BP model(GA-BP)and particle swarm algorithm optimized BP model(PSO-BP),it shows the superiority of BAS-BP wind speed forecasting model.To further improve the accuracy of wind speed prediction,a wind speed prediction model based on deep belief neural network(DBN)was established by using the powerful data feature extraction capability of the restricted Boltzmann machine(RBM)based on the shallow BAS-BP network.The model structure parameters are optimized and adjusted continuously,and the prediction performance of the model is tested by simulating the three test sample sets that are not continuous in the time dimension.Finally,based on actual wind field data for verification,the results show that on the three test samples,the MAPE,RMSE and MAE indicators of the DBN wind speed prediction model are better than the BP,GA-BP,PSO-BP and BAS-BP models,and DBN on the three test samples the model prediction accuracy fluctuates the least.This shows that the DBN wind speed prediction model has higher prediction accuracy and obvious advantages in generalization performance.