Research And Application Of Natural Gas Load Forecasting Based On Time Series Combination Model

With the continuous advancement of the transformation of the global energy consumption structure,the contradiction between supply and demand of natural gas,as a representative of clean energy,has been escalating year by year.Accurate prediction of natural gas load is of great significance to promote the rational distribution of natural gas pipeline network in China and effectively guide the peak regulation of natural gas storage.Therefore,it is necessary to find a model with high accuracy and strong robustness to predict the natural gas load,to alleviate the contradiction between supply and demand to a certain extent,and to ensure safe gas supply.The total length of a gas transmission pipeline is 2206 km,the annual gas transmission capacity is12 billion cubic meters,and there are more than 60 users along the line.This paper uses the data of 1096 groups of natural gas load of users along the pipeline on January 1,2018 solstice and December 31,2020 to carry out the following research work:First,this paper studies the characteristics of natural gas load forecasting data.The mean value interpolation method is used to fill in the missing data of natural gas load.Through the stationarity and randomness test of natural gas load,it is found that the natural gas load data is a set of non-stationary and non-pure random time series.Pearson correlation analysis method is used to verify the strong negative correlation between temperature and natural gas load.Secondly,in order to give full play to the advantages of each single forecasting model,research is carried out from two aspects: the traditional time series forecasting model and the time series forecasting model based on machine learning.In the traditional time series forecasting model,based on the "decomposition-combination" idea,the "EMD-ARIMA" combination forecasting model was established;in the machine learning time series forecasting model,the traditional PSO algorithm was improved,based on the "optimization-combination" Thought,established the "IPSO-LSTM" combined forecasting model.Finally,the load forecast of typical users A and B of this gas pipeline is carried out.900 sets of data from January 1,2018 to June 18,2020 were used as the training set,100 sets of data from June 19,2020 to September 26,2020 were used as the test set,and 96 sets of data from September 27,2020 to December 31,2020 were used as the prediction set.The prediction results were compared with ARIMA,LSTM and PSO-LSTM,and the results showed that the prediction effect of the combined model was better.When predicting user A,the RMSE of IPSO-LSTM and EMD-ARIMA are 7.10 and 11.12 respectively;when predicting user B,the RMSE of IPSOLSTM and EMD-ARIMA are both 0.30,indicating stronger robustness of IPSO-LSTM.