基于特征筛选BP神经网络的天然气需求量预测

Predicting the natural-gas demand based on feature selection and BP neural network

天然气是一种绿色高效的能源,准确预测其需求量对于天然气政策制定、生产和贸易等均具有重大意义。传统的回归分析、灰色模型等方法对天然气需求量预测仅仅考虑时间因素,无法保证预测的准确性。近年来,基于人工神经网络的天然气需求量预测被证实是一种高效准确的方法,但目前研究主要集中于算法优化,而对于天然气需求量影响因素的研究则较少。基于此,采用灰色关联度法(GRA)、平均影响值法(MIV)和主成分分析法(PCA)对天然气需求量影响因素进行了特征筛选,以简化和优化神经网络模型,提高预测的准确性和精度。研究结果表明:①三种方法均能显著降低神经网络预测误差,其中MIV法效果最佳,预测的平均相对误差从9.077%降至0.983%;②较之于传统的灰色预测模型,通过三种方法特征筛选的BP神经网络模型对2019-2025年的天然气需求量预测结果基本一致,而灰色模型预测结果偏大,表明所建立模型预测精度较高,结果可靠,可以用于对天然气需求量的中长期预测。

Natural gas is a kind of green and efficient energy.To accurately predict its demand is of great sig⁃nificance to the policy making,production,and trade of natural gas.However,in some traditional prediction meth⁃ods of both regression analysis and grey model,only time factor is considered,resulting that the prediction accu⁃racy cannot be guaranteed.In recent years,artificial neural network has been proved to be efficient and accu⁃rate in natural-gas prediction,but the current research mainly focuses on algorithm optimization,even less on the influence factors of natural-gas demand.So,BP neural network model was established to qualitatively ana⁃lyzed these factors.Then,feature selection was done on the influence factors by means of grey correlation de⁃gree method(GRA),average influence value method(MIV),and principal component analysis(PCA),so as to sim⁃plify and optimize the neural network model to improve the prediction accuracy.Results show that(1)these three methods can significantly reduce the prediction error in neural network,among which MIV method as the best one may decrease the average relative prediction error from 9.077%to 0.983%;and(2)the prediction re⁃sults of natural-gas demand from 2019 to 2025 by the BP neural network model and feature selection by the three methods are basically accordant,while those by the traditional grey model are larger,indicating that this model has high prediction accuracy and reliable prediction results and can be used for predicting the middleand long-term natural-gas demand.