基于最近邻聚类的多模型LSSVM逆控制系统

Multi-model LSSVM inverse control system based on nearest neighbor clustering algorithm

针对逆系统中非线性逆模型辨识困难以及大规模数据采用单模型回归存在精度差和计算量较大的问题,提出了一种基于最近邻聚类的多模型最小二乘支持向量机(LSSVM)逆模型辨识及控制方法。该方法首先使用最近邻聚类算法对数据集做出聚类划分,然后针对每个聚类做最小二乘支持向量回归估计,实现了对系统逆动力学模型的动态辨识。最后将辨识模型作为控制器模型,与被控对象串联,构成一个动态伪线性对象,从而使非线性对象的控制问题转换为线性对象的控制问题,仿真结果表明基于最近邻聚类的多模型LSSVM逆控制系统辨识能力强,比单模型LSSVM逆控制系统具有更优的动态跟踪性能,更好的抗干扰能力和鲁棒性。

With in mind the inverse model identification problems in inverse system method and a single model usually suffers from bad accuracy and big computation in dealing with massive dam. Therefore, a multi-model LSSVM（least square support vector machine） based on nearest neighbor clustering algorithm for nonlinear modeling was presented. Nearest neighbor clustering algorithm was used to make a partition for the data set. Then LSSVM was used to achieve regression for each cluster,realized the identification of the inverse dynamic system model. The model of controller which was the copy of identifier and the plant controlled were in series, which formed a dynamic pseudolinear system. Simulation results demonstrate that the proposed approach has a good approximate capability for inverse model, and also has better dynamic track performance, and robustness than the inverse control system based on single LSSVM.