城市天然气短期负荷预测研究

SHORT-TERM LOAD PREDICTION OF URBAN GAS

通过对城市天然气负荷变化规律和影响因素的分析,提出了一种将自组织特征映射(SOFM)网络和多层感知器(MLP)网络相混合,建立城市天然气短期负荷预测模型的方法。为使预测值不受负荷逐年变化这一趋势的影响,把负荷数据变换为特征、均值和方差的形式,利用SOFM网络预测负荷的特征,然后利用MLP网络预测负荷的日均值和方差。研究结果表明,该方法具有较高的预测精度,同时基于这种预测方法的软件包(NGPSLF)也已经成功投入商业运行,取得了满意的效果。

According to the analysis on the load-changing law and the effect factors of urban gas and combining the Self-Organizing Feature Map (SOFM) network with the Multi-Layer Perceptron (MLP) network, the method is developed to build the short-term load prediction model of urban gas. In order to make the prediction independent from the general trend of load changing year by year, the load data are transformed to the forms of profile, mean value and variance value. The SOFM network is used to predict the load profile. And the MLP network is used to predict the daily mean value and variance value. The results show the method has higher accuracy of prediction. Meanwhile, the software package of Shaanxi Natural Gas Pipeline Simulation and Load Forecasting (NGPSLF), especially for the prediction method, is available commercially. The satisfactory results have been achieved.