基于GA-SVR的煤炭需求预测模型研究

Coal demand forecasting model based on GA-SVR

在煤炭需求预测中,存在历史样本量较小和非线性强的特点,从而致使预测精度较低.将支持向量机回归(support vector regression,SVR)与遗传算法(genetic algorithm,GA)相结合,提出了适用于小样本量学习的GA-SVR煤炭需求预测模型.通过分析选取5项指标作为煤炭需求的影响变量;以历史煤炭需求与其影响变量值为学习样本,结合遗传算法确定SVR预测模型参数;实例结果表明GA-SVR模型预测精度优于BP神经网络模型.

One of the features in coal demand forecasting model is small historic sample size and strongness of non-linear,which results in lower prediction accuracy.A GA-SVR forecasting model is proposed by combining the support vector regression（SVR） and genetic algorithm（GA）.The model is proved more suitable for small sample prediction.By analyzing five selected indicators as the influence variable of coal demand,taking the historic coal demand with its influence variable value as learning samples,combined with genetic algorithm,the parameters of SVR model are determined.The result of a case shows that the prediction accuracy of GA-SVR is better than BP NET.