| As more and more people to protect a strong awareness of the living environment. Natural gas as the high-quality and clean energy become more and more popular, so building intelligent gas official website is also developing rapidly. To be able to safely and effectively use and transportation of natural gas, the gas must be able to load value of the future within a certain period of time for scientific accurate predictions based on the characteristics of the region and the variation of the gas load.Whether we can accurately predict the gas load has a direct impact on people’s safety, laying the economic interests of plan providers and gas pipelines and other issues. While in the field of load forecasting, forecasting electricity is already a relatively mature technology, but because of the presence of natural gas and electricity which is very different in terms of physical characteristics, such as storage, so the power load forecasting method cannot be applied directly to the gas load forecasting field. In recent years, as more and more scholars and experts devoted to research in the field of gas load forecasting, it has achieved some results, but there is still lack of forecasting accuracy, low efficiency, scalability and poor. In this paper, the current research by analyzing the gas properties in the region, through continuous attempts, try to find a predictive model for local gas load forecasting. After a review of various domestic and international references found that most of the scholars and experts are committed to finding a more excellent predictive model, and few studies on how to choose a more suitable prediction model of the training set. To solve this problem, and in conjunction with local gas load characteristics, this paper proposes a method to elect the smart grid classification and someday the most relevant data to be predicted from the gas load all historical values, use this data set prediction model training, found through experiments, for regression prediction model, SVM, neural network models, etc. after these traditional classification method combines the use of smart grid, in terms of prediction accuracy and prediction efficiency generally has improved to some extent. Although the prediction accuracy has improved to some extent, but still cannot meet the production error rate controlled at about 0.05 needs, so this paper also proposed the use of cross-validation, respectively(mathematics), genetic algorithm(biology) and simulated annealing algorithm(physics) on the fuzzy neural network parameters are optimized, and then combined with the smart grid classification method of gas load forecasting. Experimental results show that the combination of these three methods are basically meet production demand error rate was controlled at 0.05 or so, but differ in terms of operational efficiency, it is possible to choose a different prediction model to predict the actual situation. Finally, the use of hybrid programming technology to achieve the above prediction methodThrough the study of the impact analysis of the gas load date type found weekday, weekend holidays in the amount of gas usage and there is a clear difference, so this all the historical load data into the working data, data on weekends and national holidays Research data were predicted, resulting in more accurate predictions. |
