Short-Term Load Forecasting Of City Gas Based On PCA-WNN Model

With the urban construction and economic development in China,the city gas industry has developed rapidly.In order to properly distribute gas transmission and distribution and urban dispatch,it is necessary to predict the short-term load of city gas.Traditional prediction methods such as regression analysis,Bayesian estimation model,support vector machine,and time series model cannot meet the urban gas load forecast due to its limitations.To this end,it is necessary to develop new models and methods for urban gas load forecasting.The city gas load has the characteristics of large fluctuation,high randomness,strong irregularity,and wide influence factors.This paper focuses on the shortcomings of the existing urban gas load forecasting methods,such as low accuracy,easy to fall into local optimum,complex modeling process,long time for solving the model,and difficulty in guaranteeing the uniqueness of the solution.The factors affecting the load value were calculated and ranked by the PCA principal component analysis method,and five main influencing factors were selected.The disadvantages of BP neural network are analyzed scientifically,and the superiority of wavelet neural network as a short-term load forecasting model for urban gas is introduced.Then,the MATLAB software is used to construct the wavelet neural network prediction model:PCA-WNN hybrid model,and PCA is the main analysis.The five influencing factors selected by the method are used as the input layer of the model.The Morlet wavelet basis function is used as the hidden layer of the model,and the output layer is the predicted load value.A total of 1045 sets of data from 2011 to 2013 were used as research objects.The first 995 sets of data are used for the training model,and the last 50 sets of data are used as a comparison reference for the predicted values,and the relative error of the model is calculated to analyze the prediction accuracy.The prediction of PCA-WNN model is compared with BP neural network model,grey prediction model and GRNN generalized regression neural network.It is finally determined that the model has obvious advantages in short-term load forecasting of city gas,improving the accuracy and calculation speed of city gas load forecasting,and promoting the development of urban gas short-term load forecasting.