论文部分内容阅读
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。
In order to solve the problem of multi-batch and multi-condition chemical process, the off-line model is prone to aging failure and can not meet the real-time optimal control of industrial production, an on-line modeling method of LS-SVM based on affine propagation clustering and dynamic time warping distance is proposed. In this method, the affine propagation clustering algorithm is used to divide the working samples into different batches. Then the temporal order of the samples is considered. A time series containing the sample to be tested is taken as the query sequence, and the dynamic time warping distance The similarities between the sequences were measured. Similar sample segments were obtained from corresponding working conditions of historical batches, and training sample sets were constructed. Finally, an online prediction model was established by using least square support vector machine. The method is applied to the prediction of penicillin concentration. The simulation results show that the proposed method improves the prediction precision and generalization ability of the model.