Gas Daily Load Prediction Based On CEEMDAN-MIPCA-LSTM Model

In recent years,my country pay attention to the development of the natural gas industry,and the proportion of natural gas in primary energy is showing a continuous growth trend.In this context,accurate prediction of natural gas usage has become an urgent problem to be solved.Accurate load forecasting can provide gas suppliers with a basis for decision-making and rationally deploy the relationship between supply and demand,which has important reference significance for the healthy development of the industry.Therefore,accurate gas load forecasting is a problem that the country urgently needs to solve.This article mainly solves two problems in the forecast.The first problem is that the data dimension is too high.The gas load is affected by various factors such as weather conditions and economic development.The feature factor has a large redundancy,which causes a decrease in prediction accuracy.In order to solve this problem,the correlation analysis is added in the process of dimensionality reduction,and the mutual information coefficients of features and loads are used instead of the eigenvalues of the characteristic covariance matrix to perform dimensionality reduction operations,which removes the correlation between features while retaining the load and load.More relevant features,and establish an LSTM neural network model optimized by mutual information principal component analysis for verification.The second problem is that gas data has high complexity,and mutual interference between information affects the results of the prediction model.Therefore,this paper uses the adaptive noise integrated empirical mode decomposition(CEEMDAN)to decompose the gas load,obtain the modal characteristic information of the gas at different frequency scales,and establish the corresponding LSTM models for the different IMF components generated by the decomposition to predict,and Establish a CEEMDANLSTM model for verification.In the end,both optimization algorithms proved their respective effectiveness and improved the prediction accuracy.Finally,the two optimization algorithms are combined to form a new CEEMDANMIPCA-LSTM prediction model.The model also completes the dual extraction of load values and features at the same time,comprehensively considering the respective advantages of the two optimization algorithms,and solves the problems of too high data dimensions and more complex gas data.In the process of model prediction,the principal component factors generated by MIPCA Each IMF component decomposed by CEEMDAN is recombined into several sets of training matrices as the input data of the LSTM network to establish a prediction model.The CEEMD-MIPCA-LSTM model fully considers the influencing factors of the gas data,the inherent characteristics of the load data itself,and the characteristics of the load data on the time scale.By establishing the CEEMD-MIPCA-LSTM gas load forecasting model,and then comparing and analyzing with other models,it is finally proved that the forecasting performance of this model is better.