基于扩展核熵负载矩阵的发酵过程故障监测

Fault monitoring of fermentation process based on extended kernel entropy load matrix

中文关键词  过程监测  主元分析  多阶段  发酵过程

英文关键词  process monitoring  principal component analysis  multistage  fermentation process

中文摘要

      为有效降低多阶段发酵过程硬分类缺陷而导致的误报和漏报率, 本文提出了一种基于扩展核熵负载矩阵的阶 段划分策略. 首先, 将发酵过程的三维训练数据按批次方向展开成二维数据矩阵, 对每个时间片矩阵进行核熵成分分 析(kernel entropy component analysis, KECA)得到其主元和负载矩阵, 根据所得主元个数实现操作阶段的第1步划分; 之后将时间片矩阵添加到核熵负载矩阵当中得到扩展核熵负载矩阵, 计算各扩展负载矩阵间的相似度, 并用模糊C–均 值方法对其进行第二次阶段划分. 通过增加对体现生产过程改变的时间指标的考虑, 有效克服了硬化分的不足, 避免了 跳变点错分的情况. 最终将整个生产操作过程划分为不同的稳定阶段和过渡阶段, 并在划分的每一阶段中分别建 立KECA监测模型; 最后利用青霉素发酵仿真平台和大肠杆菌生产白介素–2数据进行实验. 实验结果表明该方法不但可 以准确地对生产过程进行阶段划分、降低误报率, 而且可以使生产过程故障监测的时间大大提前.

英文摘要

      Hard classification for multistage fermentation process and cause of the defects of false alarm and alarm failure, in order to effectively reduce the omission and the rate of false positives, this paper proposes a strategy based on extended nuclear entropy load matrix. First, the three-mention training data array of fermentation process is unfolded in batch ways, resulting in two-dimension forms. Then, kernel entropy component analysis (KECA) was done for each time slice matrix to obtain its load matrix. After that, time slice matrix was added to the nuclear load matrix of entropy, and the change of the nuclear load matrix of entropy was utilized to describe the changes of batch processes.The KECA monitoring model was established at each stage of the division after the stage of nuclear load matrix of entropy was determined by FCM algorithm. At last, the effectiveness and utility of the proposed method were validated through the simulation of fed-batch penicillin and E. coli production of interleukin-2. Results showed, the proposed method could not only divide the stage and reduce the false alarm precisely, but also detect the production difficulty more advance.