论文部分内容阅读
支持向量机建模中的一个关键和难点问题是自由参数的设置.不同于以往应用残差的简单统计量选取最佳模型的方法,本文提出通过检验模型在训练集上的拟合残差是否不含冗余信息作为选择自由参数的依据.进一步提出应用全向相关函数(omni-directional correlaton function,ODCF)检验残差信息冗余并给出应用方法,并从理论分析和数值仿真两方面给出该方法正确性的证明.在两个典型的非线性时间序列(年均太阳黑子数和Mackey-Glass数据)上进行了实验,实验结果优于相关文献记载及基于校验集方法的预测性能.
One of the key and difficult problems in SVM modeling is the setting of free parameters.Different from the method of selecting the best model from the simple statistic of residuals in the past, this paper proposes whether the residuals of the models fitted to the training set by checking the model are Which does not contain redundant information as the basis for choosing free parameters.It is further proposed to test the residual information redundancy by using the omni-directional correlaton function (ODCF) and give the application method. From the theoretical analysis and numerical simulation, The proof of the correctness of this method is given.Experiments are performed on two typical nonlinear time series (annual average sunspot number and Mackey-Glass data), and the experimental results are better than the related literature and predictive performance based on the test set method .