基于重要点多模型的不等长间歇过程弱故障检测

Weak Fault Monitoring for Uneven-length Batch Processes Based on IP-MDKPCA

为了提高不等长间歇过程弱故障检测性能,同时降低算法的计算复杂度,提出了基于重要点多模型(IP-MDKPCA)的不等长间歇过程监测方法。该方法结合核主元分析(KPCA)和时间序列模型捕捉过程动态性和非线性,分阶段单批次建模并聚类,构建多模型监测过程中的弱故障。采用重要点提取方法,不仅解决了批次数据不等长问题还大大减少了计算复杂度。将提出的方法应用于青霉素发酵过程的监控中,验证了提出方法的有效性。

In order to improveweak fault monitoring performance of uneven-length batch processes, and decrease the computational complexity of the algorithm, fault monitoring method based on IP-MDKPCA for uneven-length batch processes is proposed. The method integrates kernel principal component analysis （KPCA） and time series model to capture dynamics and nonlinearity in processes. For each stage a single batch model is developed and clustering is implemented among single batch models to monitor weak fault. The method is used to monitor uneven-length batch process by extracting important points, which not only can solve the uneven-length problem but also can greatly reduce the computational complexity. The proposed method in this paper is applied to fault detection for benchmark of fed-batch penicillin production. The effectiveness of the proposed method is verified.