Short-term Gas Load Forecasting Method Based On WT-LMD And GRU

As a clean green energy,natural gas is widely used in many industrial fields because of its economic benefits and high safety.With the development of urban gas pipeline network,gas load forecasting has become an important index for gas companies to carry out project planning and pipe network management.However,due to weather,temperature and other factors,the gas load is difficult to predict.Therefore,how to select load forecasting methods to improve the forecasting accuracy is the key to gas load forecasting.Firstly,the temporal characteristics of Shanghai gas load data are analyzed in detail,and the correlation between prediction data and historical data is determined,which provides a theoretical basis for selecting appropriate prediction algorithm.For inevitably invalid data in the process of data acquisition,several common signal analysis algorithms are compared in detail,which proves the necessity and advancement of choosing local mean decomposition(LMD)algorithm for data processing.After that,the current situation of gas load forecasting based on neural network is introduced,and the advantages of using feedback neural network(RNs)instead of feedforward neural network(FNN)for gas load forecasting are analyzed.The variants of RNNs,the model structures of LSTM and GRU are compared and the forecasting results are analyzed through experiments.The prediction accuracy of Ming GRU neural network is similar to that of LSTM network,and the convergence speed is faster.Therefore,this paper chooses GRU neural network as the prediction algorithm.Then,a combined forecasting model based on improved LMD algorithm and GRU is proposed.Newton interpolation method is used instead of moving average method to optimize LMD algorithm to solve over-smoothing problem.PF components obtained by LMD decomposition are processed by wavelet denoising to remove invalid data.The advantages of decomposition before denoising are proved by experiments.Finally,the comparison between the combination forecasting model and the single model and the concrete process of simulation experiment are analyzed in detail.In this paper,wavelet denoising is used to process noise data,and RNNs variant GRU neural network is used as prediction model.Taking Shanghai urban gas load data as experimental data,the prediction accuracy of single GRU network,LMD algorithm and GRU combination model,and improved LMD algorithm and GRU combination model are compared respectively.The experimental results show that the prediction accuracy of the proposed model is higher than that of the traditional combination model,and the prediction accuracy of the combination model is higher than that of the single model,which proves the superiority and feasibility of the proposed model.