In industrial manufacturing,most fermentation processes are inherently multiphase and uneven-length batch processes in nature.Based on different dynamic nonlinear characteristics of different fermentation phases,a new strategy is proposed by using multi-phase dynamic principal component analysis(PCA) for fermentation process monitoring.Using Gaussian mixture model(GMM) clustering arithmetic,fermentation process data are divided into several operation stages,since GMM is adopted to discriminate different operation modes.Then,run-to-run variations among different instances of a phase are synchronized by using dynamic time warping(DTW),and sub-phase dynamic PCA models are developed for every phase.Finally,the proposed method is applied to monitor both the industrial processes of fed-batch penicillin production and interleukin-2 production in recombinant E.coli.Results demonstrate that fewer false alarms and small fault detection delay are obtained and the algorithm is proved to be efficient.%针对间隙发酵过程具有多阶段、批次不等长,且过程动态非线性往往与发酵阶段密切相关等特点,提出一种基于多阶段动态主元分析（principal component analysis,PCA）的故障监测策略.该方法采用高斯混合模型（Gaussian mixture model,GMM）对过程数据进行聚类,能客观反映不同阶段操作模态的数据分布特点,可实现子阶段划分.针对各批次阶段划分后存在的不同步问题,采用动态时间错位（dynamic time warping,DTW）方法对各阶段进行轨迹同步,对同步后的子阶段建立动态PCA模型.最后以工业青霉素发酵过程和重组大肠杆菌制备白介素-2发酵过程为背景,采用多阶段动态PCA策略对其进行故障监测,发现算法能有效降低运行过程的漏报和误报率,验证了算法的有效性。
展开▼
