Prediction Of Precipitation In The Xilin River Basin By Combined Model

The precipitation is influenced by atmospheric circulation,terrain,air pressure and other environments,which have the characteristics of complex nonlinearity,so its prediction is more difficult.Early scholars have made great progress by using a single prediction model,but the accuracy is often limited.The reason for this is that using a single prediction method may have to lose some useful information in the data series.In recent years,the use of combined models,that is,a variety of single models in an appropriate way to combine to predict precipitation has become the focus of research.When using the combination model prediction,we can combine the different advantages of each single model,and the models are connected and supplemented each other and overcome their shortcomings to fully reflect the information of the data series.If the forecaster chooses only one of these forecasting methods,it is possible to discard other useful information,so it is especially important to combine multiple single models in an appropriate way.In this paper,Based on the precipitation data of the Xilin River basin,the following studies are carried out on the precipitation data of the Xilin River basin,using three single models and three combination models.The final calculation of the RMSE was 7.9229,the MAE was 62.9613,and the model fitting latage R2 was 0.9475,which reflects the validity of the SARIMA-BP-Markov combination model for the prediction of precipitation in the Schilling River basin.The main content of this article includes the following four aspects:(1)EMD decomposition of precipitation data in the Xilin River basin and timescale analysis of decomposition results,the decomposition of the data sequence are divided into high frequency,low frequency and trend items according to the variance contribution rate of the sequence.(2)According to the results of the time scale analysis,the seasonal self-regression sliding average model SARIMA,BP neural network and weighted Markov model were selected for prediction,and choose the different combinations of the three models,one of which added the prediction results of the three models and used particle group optimization algorithms to optimize the weights.The other is that the original time series is predicted by SARIMA and its residual is predicted by BP neural networks and weighted Markov,and the result will be combined to form the final prediction result.(3)According to the classification results,different EMD-SVR models are selected to predict the precipitation sequence,and the three types of results are added together.(4)The above models made precipitation prediction for 2016,and selected three commonly used indicators,average absolute error(MAE),mean square root error(RMSE)and fit superiority(R2),to make a comprehensive evaluation of the model.The best prediction model is filtered out.