Time Series Prediction And Spatial Interpolation Of Urban Wind Field Based On Machine Learning

Since the 1970 s,China's economy and urbanization have developed speedily.At the same time,problems such as urban heat islands,climate change,air pollution and lack of energy have become increasingly prominent.These issues are closely related to the temporal and spatial distribution of wind,so the temporal and spatial prediction of wind speed has become an important issue that has attracted much attention.The research is based on the observation data of the automatic observation station in Shenzhen.The research contains two parts: the time series prediction of wind speed and the spatial interpolation of wind speed.In the part of the time series prediction of wind speed,this study considers wind direction,temperature,relative humidity,dew point temperature,air pressure,sea level pressure and local change characteristics,which have complex nonlinear correlations with the time series changes of wind speed as influencing factors.And built a stack long short-term memory network to achieve short-term prediction of wind speed.In the part of spatial interpolation of wind speed,three gradient boosting tree regression models were constructed to correct the interpolation results of the inverse distance weighting method,the modified inverse distance weighting method,and the gradient inverse distance weighting method.The interpolation results,terrain features and time features of the traditional spatial interpolation method are used as the input of the gradient boosting tree regression model.The experimental results show that the mean absolute error(MAE),mean square error(MSE)and root mean square error(RMSE)of the stack long short-term memory network model are 0.431.m/s,0.339m/s and 0.582m/s are all smaller than the support vector machine regression model.At the same time,when the input only includes meteorological features,the prediction error of the model increases significantly,which shows that when the model adds local features as input,the error of wind speed time series prediction can be effectively reduced..The gradient boosting tree regression model effectively corrected the interpolation results of the traditional interpolation method.In the stratified sampling test,the range of mean square error decreases from0.835 to 2.141m/s to 0.574 to 0.860m/s,and the range of coefficient of determination was from 0.239 to 0.494 increases to 0.581 to 0.736.In the leave-one-out crossvalidation test,the gradient boosting tree regression model corrects the interpolation of some unknown regions.