Improved GM(1,1)-ARIMA-LR Model for Natural Gas Production Prediction
The study aims to improve the prediction accuracy of natural gas production with small sample.On the basis of the idea of learning from the past prediction errors and with adaptive learning factors and combined learning factors added,an integrated prediction model including GM(1,1),ARIMA and LR is constructed.The model takes the average error percentage as the evaluation index,dynamically adjusts the single model according to the change of prediction step size and the past prediction errors,and then establishes the objective programming model to dynamically weight each model.The empirical results show that the modified GM(1,1)-ARIMA-LR model can better extract the long-short dependence relationship of time series,achieving higher prediction accuracy compared to other typical models.One-,five-and eight-step predictions of the natural gas production over the last five years were made by the GM(1,1)-ARIMA-LR integrated model,with the errors being 1.187％,3.129％and 9.855％,respectively.And furthermore by the model,China's natural gas production from 2023 to 2030 was predicted.
natural gas productionARIMA modelrrey GM(1,1)modellinear regressionmultistep prediction
万方数据
维普
