基于RVM-GRNN组合模型的天然气负荷预测研究

Research on Natural Gas Load Forecasting Based on RVM-GRNN Combined Model

为了提高天然气负荷预测精度,针对不同时间段的天然气负荷周期性及非线性特点,提出一种相关向量机模型(RVM)和广义回归神经网络模型(GRNN)组合的优化模型。采用RVM对天然气负荷数据值数据进行初步建模,并用GRNN对RVM模型的残差进行非线性建模。将RVM模型、GRNN模型及RVM-GRNN组合模型对集中供热和非供热阶段的天然气负荷值分别进行预测,将组合模型分别与单一模型预测结果进行比较,并通过实际案例加以验证。实验结果表明,组合模型预测精度高于单一模型预测精度,在非供热阶段和集中供热阶段,组合模型的MAE、MSE、MAPE均小于单一模型,分别为0.1558、0.0472、0.0416和0.9597、1.6603、0.0279。除与自身单一模型进行比较外,将组合模型预测传统负荷预测模型进行比较,结果显示组合模型预测结果均优于传统预测模型。由此得出,RVMGRNN组合模型能够捕捉天然气负荷值变化规律,满足天然气负荷预测要求,可为天然气输送及管网铺设提供依据。

In order to improve the accuracy of natural gas load forecasting,according to the periodicity and nonlinearity of natural gas load in different time periods,an optimization model based on the combination of correlation vector machine model(RVM)and generalized regression neural network model(GRNN)is proposed in this paper.RVM is used to preliminarily model the natural gas load data,and GRNN is used to nonlinear model the residual of RVM model.The RVM model,the GRNN model and the first mock exam RVM-GRNN model are used to predict the natural gas load values of the central heating and non heating stages respectively.The combined models are compared with the prediction results of the single model respectively.The first mock exam shows that the first mock exam is more accurate than the single model.In the non heating stage and the central heating stage,the combined models MAE,MSE and MAPE are all less than the single model,which are 0.1558,0.0472,0.0416 and 0.9597,1.6603,0.0279 respectively.In addition to comparing the results of the first mock exam with the traditional model,the results of the combination model predict that the combined model is better than the traditional prediction model.Therefore,RVM-GRNN combined model can capture the change law of natural gas load value,meet the requirements of natural gas load prediction,and provide basis for natural gas transmission and pipe network laying.