Study On Daily Load Characteristics Analysis And Prediction Methods Of Winter Gas In Cities And Towns Of Sichuan And Chongqing

Forecasting daily gas load accurately of each terminal in urban gas network is the pre-condition for the upstream, midstream and downstream gas enterprises to work efficiently. With exact forecasting values of gas load, these companies can make reasonable sales plans, design the gas network city and realize the balance between gas supply and demand. Currently, daily gas load forecasting in winter owns important research value for the lack of gas storage facilities, tight gas supply in winter and the growing demand of users in China. However, characteristics of urban daily gas load are diverse at different time intervals in winter. Differ-ent frequency components of daily gas load are distinct. Moreover, characteristics of daily gas load of cities with different gas-utilizing structure are also distinct. To ensure the daily gas load accuracy of different cities, identification and correction of abnormal data of winter daily gas load, analysis of characteristics and its influence factors of daily gas load and establish-ment of forecasting model needs to be researched. Therefore, research work carried out in this paper is as below.(1)The methods of wavelet singularity detection and wavelet threshold de-noising are adopted respectively to identify and correct the abnormal data of winter gas daily load. Com-pared with the k nearest neighbor algorithm and the weighted correction method based on characteristic curves, the wavelet singularity detection is proved to be more accurate and fast-er in the calculation speed. Moreover, the wavelet threshold denoising method is certified to have better abnormal data correction effect and better capacity to keep the useful information of daily gas load.(2) The observational daily gas load is analyzed respectively for city D with mainly civil users and city X with mainly industrial users in Sichuan Province. For the daily gas load of the two cities in recent three years, total trends, local randomness, volatile and cyclical char-acteristics are researched, individually. The correlation analysis and partial correlation analy-sis are adopted to research the change law between daily gas load and its influencing factors of the two cities. Results show that daily gas load of city D has stronger randomness and local volatility, weaker local volatility and less similarity than city X. Meanwhile, temperature and the date type are the main influencing factors of daily gas load of the two cities. For city D, temperature has greater influence on daily load than that of city X while date type has bigger influence on daily load than that of city X. Effect of maximum daily temperature and average temperature on the daily gas load is greater than daily minimum temperature. Temperature has mutability and cumulative effect on daily gas load. When temperature drops 1 to 3? and changes between 5? and 10? or between 13? and 19?, daily gas load of city D changes obviously. For city X, when its temperature changes between 4 and 10? or between 11? and 18?, daily gas load changes obviously.When temperature drops 1 to 2?, daily gas load of city X raises larger than that of city D.(3)By means of Matlab2010b software platform, the parameters of support vector ma- chine (SVM) are optimized by the cross validation method. SVM prediction models of above two cities are established individually for the periods with temperature drop and the periods without temperature drop. Moreover, accuracy and suitability of SVM models of the two cit-ies are compared and analyzed. Results show that the timesharing forecasting model based on SVM has higher accuracy and quicker calculation speed than that without timesharing. Its mean absolute percentage error is between 2% to 3.8%. The timesharing forecasting model based on SVM model is proved to have strong approximation ability and adaptability and can be applied to winter daily gas load forecasting in cities with distinct gas-utilizing structure. Meanwhile, due to large variation of daily gas load during the Spring Festival of city X, time step length of prediction is recommended to be set as 2 to 4 days to improve the prediction accuracy.(4)By means of wavelet transform, winter daily gas load of the two cities is decomposed into several high frequency components and low frequency components, in which high fre-quency component reflects random fluctuations and detail information of the daily gas load while the low-frequency component reflects the change trend of daily gas load. Two combina-tion schemes are raised respectively to set up the frequency-division combination model of daily gas load forecasting. Combination scheme 1 refers to adopting the time series model to predict the low-frequency component and using the SVM model to forecast high frequency components. Combination scheme 2 means using the SVM model predicting the low-frequency component and adopting the time series model to forecast high frequency components. Analysis results show that forecasting accuracy of the combination scheme 2 is higher than scheme 1 as well as the prediction model without frequency-division. Moreover, the overall forecasting accuracy of frequency-division combination model is higher than that of the timesharing forecasting model. However, it is weaker to forecast daily gas load during the Spring Festival for city X. Therefore, the SVM forecasting model is recommended to forecast daily gas load during the Spring Festival period of city X. And frequency-division combination model is recommended to forecast daily load during other periods of city X. For city D, the frequency-division combination model is the recommended method.