摘要
针对一类非线性系统的控制问题,结合基于仿射传播聚类的最小二乘支持向量机(LS-SVM)多模型建模算法与PSO优化算法,提出新的多模型预测控制算法.采用仿射传播聚类算法对历史样本数据进行聚类,得到各个类的训练样本数据;利用LS-SVM对各个类分别建立子模型,采用网格搜索和交叉验证为各子模型找到合适的模型参数,将所建立的子模型作为预测控制算法的预测模型.在滚动优化时,计算当前控制量与各聚类中心的欧氏距离,选择相应的子模型计算未来时刻模型的预测输出,计算得到参考轨迹.建立优化问题的目标函数,采用PSO算法优化求解得到系统的最优控制量作用于对象.将提出的算法在某芳烃异构化过程中进行仿真试验,分别采用提出的算法以及单模型预测控制算法、基于k均值和BP神经网络的多模型预测控制算法进行仿真.结果表明,采用提出的多模型预测控制算法可以获得更好的控制性能.
A new multi-model predictive control algorithm, integrating affinity propagation （AP） clustering and least squares support vector machines （LS-SVM） modeling technology with particle swarm optimization （PSO） algorithm, was presented in order to enhance the control property of a nonlinear system. Samples were clustered into several classes by AP algorithm and the training data of every class were obtained. Then corresponding sub-models were constructed as the predictive model by LS-SVM whose parameters were optimized through grid-search and cross-validation method. In rolling optimization stage, the sub-model was selected by computing the Euclidean distances between current control variable and each center of clustering, and corresponding predictive outputs were generated. Then the optimization problem was constructed according to the predictive outputs and reference trajectory. The optimal control values were obtained by the rolling PSO optimization algorithm. The presented algorithm was applied in the simulation of aromatics isomerization process. The presented algorithm, single model based predictive control algorithm and multi-models predictive control algorithm based on k-means and BP neural network were applied, respectively. Results show the better performance of the proposed multi-model predictive control algorithm.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2013年第10期1741-1746,共6页
Journal of Zhejiang University：Engineering Science
基金
国家自然科学基金资助项目(61203072
61203133)
江苏省六大人才高峰项目
工业控制技术国家重点实验室开放课题资助项目(ICT1234)
