| The current numerical weather prediction model is the main method of weather forecasting,but due to the problems of the numerical weather prediction model system itself and the error of the input meteorological field,it is inevitable that there are some errors in the prediction results.It is necessary to revise the numerical model prediction results to achieve better accuracy,and researchers at home and abroad have done a lot of work on weather forecast revision.This paper summarizes the work and algorithms used by researchers at home and abroad on weather forecast revision,and proposes a reconstruction algorithm CD-XGBoost(Clustering and Double XGBoost),based on XGBoost algorithm according to the current research on weather forecast correction based on machine learning approach.This algorithm includes three main improvement directions: firstly,it puts forward the idea of clustering by using the correlation between weather elements and revised elements,and through the training of the inter-cluster stations based on different machine learning approaches,the revision results of the machine learning model to the wind speed is improved;Secondly,the algorithm highlights the influence of spatial factors on the prediction of meteorological elements,and selects the forecast elements of K adjacent forecast grid points of meteorological observation stations to construct data sets.Compared with the traditional interpolation method,the revision method takes more account of the spatial factors between the station and the grid points.Thirdly,the ECWMF numerical weather forecasting system has two starting time points,using the initial field data of UTC00:00 and UTC12:00respectively.According to this characteristic,this paper proposes a two-model linear weighted summation model based on XGBoost algorithm to further revise the prediction system error to improve the reliability of the prediction results.In the simulation experiment of this paper,the three-hour observation data of2552 meteorological stations in China and the three-hour prediction data of(ECWMF)numerical model of European medium-range Meteorological Forecast Center are used to correct the daily maximum temperature,daily minimum temperature and 10-meter surface wind speed predicted by ECWMF.In the experiment,grid search and cross-validation are used to adjust the parameters of the basic model to improve the fitting degree of the model,and the root mean square error and prediction accuracy are used to evaluate the model.In the comparative simulation experiment,in view of the stage of the improvement of the algorithm in this paper,the results before and after the improvement of the algorithm in each stage are compared,which proves that the improvement point proposed in this paper is effective.On the whole,the prediction results of each algorithm basically increase with the length of the forecast time,and the prediction accuracy decreases as well.At the same time,the CD-XGBoost algorithm proposed in this paper is obviously better than other correction algorithms in the RMSE and accuracy evaluation of temperature and wind speed prediction.The RMSE value of the daily maximum temperature predicted by the algorithm used in this paper is 1.32?,and the accuracy is 83.12%;The RMSE value of predicting aging for seven days is 2.13?,and the accuracy is 68.17%;The accuracy rate is increased by 12.8%.The daily minimum temperature RMSE of one-day prediction aging is 0.77?,and the accuracy is 92.46%;The RMSE value of7-day aging prediction is 1.63?,and the accuracy is 75.38%;It is 10.9% higher than the average forecast accuracy before the improvement.The RMSE value of the 10-meter wind speed during the 3-hour forecast time limit is 0.68 m strokes,and the prediction accuracy is higher than 85%;The wind speed RMSE value of 168-hour forecast is 1.41 m/s,and the prediction accuracy is higher than 60%;The average forecast accuracy is mentioned at 12.3%.The experimental results show that this study has a good application prospect. |
PDF Full Text Request