Research On Short-term Gas-load Forecasting Based On Combination Methods

With the development of intelligent ’citys construction in Shanghai, the natural gaspipeline network becomes more and more intelligent. Wc should grasp thecharacteristics and changes of gas load, so as to make the gas system run in highefficiency and scientific management.Gas load forecasting is the most flindamental and critical technology in theconstruction of gas intelligent network, which is ifjll of creativities, like power load.So far many researchers have studied it for years and also got some results, but a fewproblems are still there, like low forecasting accuracy，low forecasting efifciency. Forits current situation, this paper aims to ifnd the most proper way to forecast the gasload. Firstly, it makes a study in input variable, data preprocessing.etc. usingtechnologies including data mining, correlation analysis and partial correlationanalysis. Then uses regression techniques, neural network, wavelet analysis (supportvector machine)and a combination method to analysis the intelligent forecasting gassystem. Gas load forecasting is a complex nonlinear problem. With a single predictionmethod, it is difficult to meet the authenticity and accuracy of the load forecasting.Atfer searching the literatures and comparing the advantages and disadvantages ofeach method, combined with actual data，it is ifnd out that artiifcial neural networkforecasting is the most suitable method for the short-term prediction of the gas load.This paper introduces the basic principles of the artificial neural network, thealgorithm and prediction steps with details and uses the gas load data of Shanghai totrain the artificial neural network. Then builds different models and predicts as needed.The experimental results have shown that the neural network is more feasible andaccurate for short-term gas load forecasting. Atfer combined with the wavelet analysis theory and the improved genetic algorithm, we ifnd that both of the two methods cangreatly improved the lack of the neural network and make the forecasting results moreaccurate. The combined the wavelet and neural network method is the main methodwe used in this paper to predict gas load. The forecasting results have shown that theprediction accuracy is higher than any single method and can get more satisfactoryresults, and demonstrate the combination method is feasibility and effectiveness. Inthis paper, we also use residuals sequence to training neural network，this is a novelalgorithm. Experimental show that this method can get higher accuracy result. Studieshave found that load between workday and weekend has significant difference, and inholiday this difference is more obvious, so in this paper we adopt differentexpeirmental program to forecast and this operational can get more accurate results.