Research On Temperature Time Series Prediction Based On SARIMA-LSTM Combined Model

Time series data exist widely in every field of the objective world.In recent years,with the rapid development of computer technology and statistical analysis methods,the methods of time series analysis are constantly updated,and more and more scholars begin to pay attention to the prediction of time series.The choice of which model to use in time series analysis and forecast effect still depends on the characteristics of data.As an important data in national economic construction,scientific analysis and prediction of meteorological time series can provide important theoretical basis for meteorological work.However,meteorological time series are affected by many external factors,which is a complex problem.Temperature prediction is especially important in the meteorological industry,which is closely related to people’s daily life.First of all,in order to improve the prediction accuracy of the temperature time series model,the linear trend,seasonality and non-linearity of the temperature series are analyzed,in this thesis,two temperature time series forecasting models based on STL(Seasonal-Trend decomposition procedure based on Loess)decomposition method are proposed: SARIMA temperature forecasting model and LSTM temperature forecasting model.Second,in the model algorithm,the STL decomposition method is used to decompose the temperature time series into linear trend term series,seasonal term series and nonlinear residual term series,then SARIMA model and LSTM model are used to predict the temperature decomposition series.The experimental results show that the SARIMA model based on STL decomposition method is effective in predicting the trend and season of air temperature.The LSTM model based on STL decomposition method is more effective in predicting residual term.At last,in view of the advantages and disadvantages of SARIMA and LSTM in decomposing different temperature signals,a combined model SARIMA-LSTM is proposed based on the two single models,furthermore,the decomposition sequences can be well used to achieve the goal of temperature prediction.Compared with other classical models and LSTM-ARIMA model,the combination model improves the prediction accuracy and shows a better temperature prediction effect.