Natural Gas Pipeline Network Load Forecasting Based On Support Vector Machines

As a clean and high effective energy, natural gas has been attached much importance by every walk of life. The increase of requirement brings much pressure to optimal dispatch of pipeline networks, so in order to dispatch precisely and apace, advanced methods be needed urgently. The nicety of load forecast is not only the base of optimal dispatch of pipeline networks but also has much important meaning to insure gas quantity, equipment maintain service and the end gas storage of pipeline.What the key of the pipeline networks load forecast is forecast technique, that is how to analyse load peculiarities , set up reasonable forecast model, design effective arithmetic so as to get accurate result. Support Vector Machine-SVM is a new-style Machine Learning method, which is based on Vapnik-Chervonenkis Theory of Statistical Learning Theory and Structural Risk Minimization theory. SVM can solve some practical problems with low sample size, non-linearity, high-dimension of feature space and local minimization. Being SVM has the merit of fast speed and high generalization, now it has become a powerful tool to over come over learning and "dimension calamity".This paper generalize several normal forecast methods such as Regression prediction, Level Moving, Time series, Neural Network, Gray System arithmetic, expert system, wavelet prediction and combining forecast in pipeline networks, then analyses their merits and disadvantages. Through analyses the load features and affected factors of load, then sets up a model by SVM, in which the inputs are holiday property, the load before the wanted day's and the load at the same day's before one week. This model avoid complex because of too much independent variables. Finally compared with BP network arithmetic, Time serial arithmetic and Gray System arithmetic.This paper use Matlab6.5 as a tool to programme in order to validate the research fruits. The examples indicate that the forecast model based on SVM can get a satisfied result, which provide creditable theory aegis for safe and economical running of pipeline networks.