基于仿射传播聚类和动态时间弯曲距离的LS-SVM在线建模方法  

Online modeling based on LS-SVM with affinity propagation clustering and dynamic time warping distance

作　　者：熊伟丽[1] 张乾[2] 徐保国[1,2] 

机构地区：[1]轻工过程先进控制教育部重点实验室,江南大学,江苏无锡214122 [2]江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机与应用化学》2015年第6期702-706,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金项目(21206053,21276111);江苏高校优势学科建设工程资助项目(PAPD);江苏省产学研项目(BY2014023-27)

摘　　要：针对多批次多工况化工过程,离线模型易老化失效和不易满足工业生产的实时优化控制问题,提出一种基于仿射传播聚类和动态时间弯曲距离的LS-SVM在线建模方法。该方法首先利用仿射传播聚类算法对各批次样本进行工况划分,再考虑样本间的时间有序性,由包含待测样本的一段时间序列作为查询序列,并以动态时间弯曲距离来衡量序列间的相似情况,从各历史批次相应的工况阶段获取相似样本片段,构建训练样本集,最后采用最小二乘支持向量机建立在线预测模型。将该方法用于青霉素浓度预测中,仿真研究表明,所提方法提高了建模预测精度和泛化能力。Many chemical processes have the feature of multiple batches and several working conditions. The offline model is difficult to satisfy the real-time control and optimization. A new online modeling method based on affinity propagation clustering algorithm and dynamic time warping distance is proposed in this paper. Firstly, affinity propagation clustering algorithm is used to cluster samples into different working conditions, and dynamic time warping distance is used to measure the similarity between the sample sequences. Then, using time series as a query sequence, to find similar sample clips from the corresponding working conditions and establish online training sample set. Finally, using the least squares support vector machines to build online prediction model. Applying this method to predict the concentration of penicillin, Simulation results show that the proposed method has higher precision and better generalization ability.

关 键 词：仿射传播聚类算法 动态时间弯曲距离 最小二乘支持向量机 在线建模 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]