Z地区天然气管网平衡差影响因素分析与天然气需求短期预测

Analysis of the Factors Influencing the Balance Difference of Natural Gas Pipeline Net⁃work in Z Area and the Short-term Prediction of Natural Gas Demand

为掌握管网进销平衡差波动规律,提前做好管网管存合理分配,基于Z地区天然气管网进销特征,结合近3年管网进销平衡差动态资料,从季节性、日期类型、天气温度等因素入手,分析了Z地区管网进销平衡差的影响因素,并采用方差分析方法对各影响因素进行综合评价,量化分析各因素对管网平衡差的影响程度,结果发现,冬季(供暖季)、节假日对于管网进销平衡差有显著的影响。根据方差分析结果,建立了关于管网销气量的神经网络短期预测模型。用BP、Elman、RBF神经网络模型预测日常管网销气,得到2020年3月整体销气预测值的平均相对误差分别为1.95%、1.51%、1.45%,均方根误差分别为781.34×104、703.65×104、655.31×104 m3。相比其他两种模型,RBF神经网络预测值更加贴近实际销气量,具有良好的预测结果。RBF神经网络模型预测重大节日管网销气,所得误差值稍大于日常情况下的预测值,但误差仍在合理的范围之内,证明该模型能够准确地预测重大节日下管网销气量,为Z地区管网调度运行、工况调整提供数据支持。

In order to grasp the fluctuation trends of the balance difference between the input and out-put,make a reasonable allocation of the pipe network storage in advance,based on the characteristics of the gas pipeline network in Z area,combined with the data of the balance difference between the in-put and output in recent three years,this paper analyzes the influencing factors such as seasonality,date type,weather and temperature,uses the method of variance analysis to comprehensively evaluate the influencing factors,and quantitatively analyzes the influence degree of each factor.The result shows that winter(heating season)and holidays have significant influences on the balance difference of pipeline net-work.According to the results of variance analysis,short-term prediction models of gas demand in the pipeline network are established by using neural network method:BP,Elman,RBF neural network prediction model of gas sales are used in working days,the average relative error of the prediction value of the overall gas sales in March of 2020 is 1.95%,1.51%and 1.45%,and the root mean square error is 7.8134,7.0365 and 6.5531 million cubic meters,respectively.Compared with the other two mod-els,the predicted value of RBF neural network is the closest to the actual sales volume and has good prediction results.RBF neural network prediction model of gas sales is used in major festivals,the error value of the model is slightly larger than that of the prediction model in daily situation,and the error is within a reasonable range,which proves that the model can accurately predict the gas consumption of pipe network in major festivals,providing data support for Z area network dispatching operation and condition adjustment.