Temperature Prediction And Early Warning Analysis Of High Temperature And Heat Wave Based On The Combination Of Time Series Model And XGBoost

The increasing of temperature year by year and the resulting drought,high temperature and heat wave and other extreme weather pose a serious threat to people's production and life.However,the prediction of temperature and the early warning of high temperature and heat wave are still in a stage of urgent development in practice.Based on the hourly historical data of Haikou Meteorological Automatic Station in Hainan Province in recent 5 years,this paper studied the temperature of Haikou Meteorological Station by using the combined model of time series model and XGBoost model,which mainly included four aspects: hourly temperature prediction,daily maximum temperature prediction,high temperature daily prediction and high temperature and heat wave warning.(1)In the prediction research of hourly temperature and daily maximum temperature,this paper put forward combination models based on time series model and XGBoost(SARIMA-XGBoost,ARIMA-XGBoost,Prophet-XGBoost),and compared them with the univariate time series model(Exponential smoothing,ARIMA,SARIMA and Prophet)and three representative machine learning algorithms(K-Nearest Neighbor,Random forest and Deep Neural Networks).It was found that the SARIMA-XGBoost model had the best performance on the test set(MAE: 0.85,RMSE: 1.09,MAPE: 2.90%).In the forecast of daily maximum temperature,the Prophet-XGBoost combination model performed best on the test set(MAE: 1.04,RMSE: 1.33,MAPE: 3.04%).(2)In the prediction research of high temperature days,this paper started from two research ideas.First,it discriminated high temperature days based on the prediction results of daily maximum temperature and the classification accuracy of high temperature days was90.3% in the one-month test set.Secondly,based on the process of data reconstruction and the stacking fusion model,this paper innovatively used the existing meteorological data to classify and predict the high temperature days,and the prediction accuracy on the test set of one month had reached 67.7%.Finally,based on the optimal prediction results of high temperature days,further discriminant prediction of high temperature and heat wave was made,and the accuracy reached 87.1%.In summary,this paper established a combination model of time series model and XGBoost to predict temperature and obtained better results,and established a stacking fusion model to improve the accuracy of classification prediction.In addition,this paper innovatively used meteorological data directly to classify and forecast high temperature days.