Gas Load Prediction Research Based On The Svm Model Portfolio

Gas load forecasting is the base of gas pipeline network systemcontrol and operation, also is the market basis, the predicted data is animportant basis to ensure the safety of natural gas pipeline network andthe rationality of network scheduling. And accuracy, real-time, reliability,intelligence for gas load forecasting system is the urgent problem to besolved.This paper firstly analyzes the characteristics of gas load data,then in-depth understanding of main factors influencing load change,to determine the input samples of forecast model providing a basis.Secondly, through the data pretreatment, the noisy data of gas loadsequence is corrected to be smoothing data sequence, so as toimprove the forecast model precision of training and predicting. Finally,a series of experiments and analysis,(1)compared BP neural networkwith RBF-SVM prediction accuracy, proved that SVM is better than BPneural network model in the gas load forecasting.(2) Experiment on theselection of the SVM kernel parameters and kernel function, andcomparison which is the best one, finding that the genetic algorithm isbetter than the method of cross validation, and the chaos geneticalgorithm can overcome the shortcoming of traditional geneticalgorithm, such as premature convergence, eventually finding that theperformance of Wavelet kernel SVM(Wv-SVM) is better than RBF kernelSVM(RBF-SVM). Through a series of experiments above, obtaining aconclusion that CGA Wv-SVM this combination model acquires thehighest accuracy.