面向多输入输出系统的支持向量机回归

Support vector regression algorithm for multi-input multi-output systems

目前支持向量机回归算法只适用于单输入单输出系统,在处理复杂多变量系统时多采用构造一系列单变量支持向量机模型的方法,增加了模型与控制的复杂性。针对这一问题,提出了面向多输入多输出系统的支持向量机线性回归和非线性回归方法,两种方法的不同在于映射到高维空间后的运算方法不同。与单变量支持向量机回归建模相比,采用该方法建立的多变量系统模型结构较为简单,且在小样本数的情况下,模型训练速度较快。最后将该方法应用到甲基丙烯酸甲脂的间歇聚合反应过程中,仿真结果表明该方法的有效性。

Support vector regression(SVR) algorithms are normally only used for single-output systems.Several SVR models were evaluated to identify one appropriate for multi-input multi-output systems,which requires a much more complex control system.The multi-SVR algorithm includes linear and nonlinear regression algorithms for different operations in the high-dimension space.The multi-SVR method has a less complex structure and faster training speed with less samples than a single output SVR model.Simulations of an MMA batch reactor are used to demonstrate the value of the multi-SVR method.