Abstract:

      Long-term natural gas load forecasting can solve the problem of the imbalance between supply and demand of city gas and provide assistance for the city gas company's management and running. In order to improve the accuracy of predicting the longterm natural gas load，a forecasting model of natural gas longterm load was built based on SVM-GA(Support Vector MachinesGenetic Algorithm). The relevant factors influencing natural gas consumption was analyzed and determined. In order to improve prediction accuracy， the penalty factor c and the kernel parameter g of support vector machines were optimized using genetic algorithm and cross validation methods. Optimized parameters were inputted support vector machines model and long-term natural gas load forecasting was made. In a case study from a certain city，a comparative analysis was made of the forecasting results among SVM-GA，SVM and crossvalidation method combined prediction model and BP(Back Propagation) neural networks. The forecasting model based on SVM-GA was validated with a high prediction accuracy and the resulted relative mean square error，normalization mean square error，normalization absolute square error，normalization rootmean square error， maximum absolute error resulted from the SVM-GA were lower than those from SVM and crossvalidation method combined prediction model or BP neural networks by 0.58%，3.98%，2.99%，4.58%，8.64% and 6.13%，26.28%，19.71%，21.09%，31.48%. Therefore，the support vector machine and genetic algorithm combined model can accurately predict the long-term natural gas load.

Key words: Natural gas long-term load,    , SVM,    , BP neural networks,    , Genetic algorithm,    , Cross validation,    , Forecast,    , Accuracy

摘要：         天然气长期负荷预测能够解决城市燃气供需不平衡的问题,为城市燃气公司的管理运行提供帮助。为了提高天然气长期负荷的预测精度,提出了基于SVM-GA(SupportVectorMachines-GeneticAlgorithm)的天然气长期负荷预测模型。分析确定影响天然气用气量的相关因素,应用遗传算法和交叉验证方法分别对支持向量机模型的惩罚因子c 及核函数参数g 进行优化,以期提高支持向量机模型的预测精度,将优化后的参数输入支持向量机模型中,进行天然气长期负荷预测。以某省实际的样本数据为例,将SVM-GA模型的预测结果与SVM 和交叉验证法结合模型及BP(BackPropagation)神经网络模型的预测结果进行比较分析。结果表明,SVM-GA 预测模型分别比SVM 和交叉验证法结合预测模型和BP神经网络模型在衡量预测精度的相对均方误差、归一化均方误差、归一化绝对平方误差、归一化均方根误差、最大绝对误差五个指标分别高0.58%、3.98%、2.99%、4.58%、8.64%和6.13%、26.28%、19.71%、21.09%、31.48%。因此支持向量机与遗传算法相结合的模型能够较准确地预测天然气长期负荷。

关键词: 天然气长期负荷,    , SVM,    , BP神经网络,    ,  遗传算法,    ,  交叉验证法,    ,  预测,    ,  精度