Research On Time Series Prediction Method Based On Hybrid Model

The time series data in reality has complex characteristics,and the traditional single model has great limitations in prediction.In order to further improve the prediction accuracy of time series,two kinds of time series prediction methods based on hybrid model are proposed in this paper.One is the hybrid prediction method based on residual,and the other is the hybrid prediction method based on decomposition and synthesis.The main research contents are as follows:In order to improve the prediction accuracy and make full use of the useful information in the residual data,a hybrid prediction model based on the residual ARIMA-LSTM is proposed.First,ARIMA model is built on the data and capture its linear components.The prediction of ARIAM is completed by checking the data stationarity and the order of the model.The residual is obtained,the residual is subjected to white noise test,and then the LSTM model prediction is established for the residual.Finally,the two partial prediction values are accumulated to obtain the final predicted value.Experiments show that the prediction model based on residuals is more accurate than the single model.For the problem that the time series data is too complicated,a hybrid prediction model based on decomposition and synthesis is proposed.The time series is decomposed by DWT,EMD and MEEMD methods.In order to establish a suitable model for the decomposed subsequences,the chaotic characteristics of the decomposed data are judged.A nonlinear model is established for subsequences with chaotic properties,a linear model is established for subsequences without chaotic properties,and the two prediction values are summed to obtain the final prediction results.For the problem that the calculation amount becomes larger after decomposition,the sample entropy(SE)method is used to combine the decomposed subsequences,which greatly reduces the computational complexity.Finally,the ELman model prediction is established for the subsequence.The experimental results show that the prediction accuracy of the model is higher than that of the comparison model.