The Research On Shanghai Gas Load Forecasting

As a green energy resource, natural gas is the development direction of the gas in the city. Increasing the natural gas proportion in energy consumption structure is not only good for energy conservation and emission reduction, but also can maintain the sustainable development of economy and the society. Shanghai one of the first cities to use natural gas which is greatly promoted now, especially the joining up of the West to East natural gas transmission project has really promoted the rapid development of the natural gas in Shanghai. In order to realize the efficient operation of gas supply system, optimized dispatching and scientific management, gas load predication, the decision base, is obviously very important.Through the analysis of load law of Shanghai gas system; this paper gives undivided attention to the precise predication of gas load. It did an intensive study of different popular intelligent forecasting techniques, such as Data Mining(DM), Artificial Neural Network(ANN), Support vector machine (SVM), Particle Swarm Optimization (PSO) optimization algorithm and so on. It also explores a combination of two predictive models.The first one is combine the rough set and BP network together by using the data preprocessing which happens through rough set to noise, redundancies and irrelevant data. Then train and predicate it in BP network as the input variable, which improves a lot than the pure BP network.The second one is daily load forecasting model, which synthesizing SOFM and SVR predication technique. In this model, we use SOFM to cluster the training data samples, and every data group has similar features, and then set up forecasting model by using SVR to the data after clustering. We adopt the CPSO optimization algorithm on the parameter of SVR kernel function. Besides, we use the data mining to the pretreatment of the load data in the past; revise the discrete node to make it more obvious on the regularity of the gas load; choose the data based on the feature of filter, and the factor which has considerable influence is the input variable of SOFM. The results indicate that this model made a great improvement, not only in training time, but also the forecast accuracy.