City Gas Load Forecasting Based On Bi-reservior Esn And Improved Rbfnn

In the context of sustainable development,natural gas,a clean energy,is in line with the theme of “building a smart green city”.City gas load forecasting plays an important role in the operation and scheduling of city gas pipeline network systems.It is of great significance to the gas supply system,pipeline construction and optimal dispatch of gas companies.In consideration of the periodic and nonlinear characteristics of gas load data and the limitations of single model,this paper used a combined prediction model based on bi-reservoir ESN and improved RBF neural network.The data used in this paper was the gas load data from 2005 to 2014 in a certain area of Shanghai.Before using the data for forecasting,data preprocessing was done first.The bi-reservoir Echo State Network(BRESN)was used for forecasting.The load sequence and parameter sequence were respectively modeled,and the output of the two reservoirs was integrated to get prediction results.Thus,prediction accuracy was improved.Fruit Fly Optimization Algorithm(FOA),with a smaller amount of calculation,was selected to optimize the four parameters(scale,spectral radius,scale of expansion and sparsity).These parameters determined the performance of BRESN.As another algorithm of the combined model,the improved RBF network takes the output of the optimized BRESN network as input.In other words,the BRESN and RBF neural networks were serialized.BRESN was used as a preliminary prediction.RBF was used as a residual correction.Therefore,the limitations of a single model were compensated.In this paper,a hybrid coding method of the binary coding and real coding were performed simultaneously on each node.The differential evolution(DE)algorithm was combined with the gradient descent algorithm,and the structure and parameters of the RBFNN were simultaneously optimized.Optimization enhanced the local search capability of the algorithm and speeded up the convergence.The last part of the paper is experiment,in order to illustrate the performance of the model,the model was compared with forecasting models based on SVM,BPNN,RBFNN,BRESN and BRESN-RBF by calculating their MAE,RMSE,MAPE,MSE.The experimental results showed that the prediction accuracy of this model was higher.