Research On Temperature Forecast Of Beijing Based On ARIMA-SVR Combinational Model

Meteorological time series are important data for the construction of the national economy,their rational use and prediction through scientific algorithms can provide an important theoretical basis for the meteorological industry to prevent meteorological disasters.Due to the influence of various external factors,the meteorological time series is an intricate system.The temperature prediction is an extremely important task in the meteorological industry and is closely related to people's lives.In this paper,by constructing the ARIMA model and support vector machine regression(SVR)model to predict the daily maximum temperature in Beijing in 2016,based on the advantages and disadvantages of the two models,combined with the idea of the combined model,further construct the five combined model of different algorithms,through the comparison of the performance of the seven models,selects the optimal model for short-term temperature prediction,and draws the following conclusions.By decomposing the trend of the temperature series,it is found that there are obvious seasonal effects on Beijing's temperature series,and a cyclic trend with a half-month cycle.The fitting effect of the traditional ARIMA model is not good.The prediction accuracy of the traditional ARIMA model has been improved to a certain extent.However,the prediction results on the next 10 days are still higher than true value by about 1.5 ?;by cross-validation and grid search method to optimize the parameters,the accuracy of the SVR model is high,and the change trend and size of the predicted values of the next 10 days are the same as the true value Very close.The SVR model has MAE = 0.756 and MAPE = 0.197.Compared with the SARIMA model,MAPE is improved by 59%;the SARIMA model and the SVR model perform well in fitting the linear and nonlinear characteristics of the sequence,respectively.However,since a single model can only reflect part of the characteristics of the sequence,a combinational model is introduced to establish a tandem ARIMA-SVR combinational model,a particle swarm optimization algorithm is used to optimize the weights of the two models,and a parallel combination model based on the equal weight method,Simple weighted average method,reciprocal variance method and other combinational models for assigning weights.Taking MAPE as a metric,through comparison,the fitting effect of the combined model of ARIMA-SVR model and particle swarm optimization algorithm is relatively good,MAPE is 0.106,0.12,respectively.Compared with the SVR model,it is increased by 46% and 39% respectively.But the fitting effect of the other three combinational models is not better than the SVR model.Through empirical analysis,it is proved that the ARIMA-SVR model is feasible and advantageous in the short-term prediction of air temperature series.This model integrates the respective advantages of two single models,reflects the changing characteristics of the data,and can provide a theoretical basis for meteorological prediction.