基于遗传算法的SVM-AR改进模型与应用

Improved model and application of SVM-AR based on genetic algorithm

为提升河流流量的预测精度,将支持向量机与AR进行耦合,并构造三核混合核函数的流量预测支持向量机模型。以渭河流域的月径流量为例,首先,通过时间序列分析,将渭河流域的径流序列划分为趋势序列、季节序列和随机波动序列,然后利用AR模型构造适用于支持向量机算法的数据集,并将数据集按4∶1划分为训练集和检验集;其次,利用线性组合构造由多项式核函数、径向基核函数与Sigmoid核函数构成的三核混合核函数,在训练集上,采用遗传算法确定相关参数,随后在检验集上进行预测。结果表明:遗传算法确定参数会带来较大的不确定性,导致结果差异较大,从而着重讨论遗传算法带来的参数不确定性;通过函数构造与统计分析,给出三核混合核函数参数选择的一般性方法与流程,并进行验证,该参数选取方法能够降低遗传算法的不确定性,得到精度较高的流量预测结果,预测流量与实际流量的均方误差从150左右降低到130左右。

In order to improve the prediction accuracy of river flow,the support vector machine(SVM)and the AR model are coupled to construct the SVM model for the river flow prediction with three-core hybrid kernel function.Taking the monthly runoff in the Weihe River Basin as an example,firstly through the time series analysis,the runoff data of Weihe River Basin was divided into trend data,seasonal data and random fluctuation data.A data set suitable for support vector machine algorithms was constructed with the AR model,and then was divided into the training set and the test set by 4∶1.Secondly,with the linear combination,this study constructed a three-core hybrid kernel function composed of polynomial kernel function,radial basis kernel function and Sigmoid kernel function.On the training set,the genetic algorithm was used to determine the relevant parameters,and predictions were conducted on the test set.It was found that when the genetic algorithm is used to determine the parameters,it will bring greater uncertainty with greater differences in results,thus it need more discussions on the parameter uncertainty brought by genetic algorithm.Through the function construction and statistical analysis,the general method and process of parameter selection of the three-core hybrid kernel function are given and verified.With this method,the uncertainty of the genetic algorithm can be reduced on the test set,and a more accurate flow prediction result can be achieved.The mean square error between the predicted flow rate and the actual flow rate is reduced from about 150 to about 130.