基于GMM和贝叶斯推理的多模态过程运行状态评价

Operation performance assessment for multimode processes based on GMM and Bayesian inference

中文关键词  多模态过程  运行状态评价  非优原因追溯  高斯混合模型  贝叶斯理论

英文关键词  multimode process  operating performance assessment  nonoptimal cause identification  Gaussian mixture model (GMM)  Bayesian inference

中文摘要

      为使综合经济效益最大化, 生产过程应保持在最优运行状态等级. 针对多模态过程运行状态等级优劣判断 问题, 提出一种运行状态等级评价方法. 该方法对同一运行状态等级的多模态数据建立一个高斯混合模型 (Gaussian mixture model, GMM), 确保特征提取的准确性, 避免模态划分问题. 至于在线评价策略, 本文采用贝叶斯 推理, 确定当前运行状态属于各等级的后验概率. 并引入滑动窗口, 判定当前运行状态等级, 有效解决多模态过程运 行状态在线评价问题. 针对“非优”运行状态, 本文提出一种基于变量偏导数的贡献计算方法, 对导致过程运行状态 等级“非优”的原因变量进行追溯. 最后, 通过田纳西–伊斯曼(Tennessee–Eastman, TE) 过程验证所提方法的有效性.

英文摘要

      To maximize the comprehensive economic benefits of enterprises, the production process ought to be kept in the optimal operating performance grade. To solve the problem of process state judgement for multimode processes, a novel operation performance assessing approach is proposed in this paper. One Gaussian mixture model (GMM) is established for a same running grade with multi modes in this article, ensuring the precision of feature extraction and avoiding mode division. As to online evaluation strategy, Bayesian inference is applied to calculate the Posterior probability of the current performance belonging to each grade. Sliding window is then introduced to help determine the running state. The proposed method turns to be an effective solution to the multi-modal process operating performance optimality online assessment. A novel variable contribution calculation technique is subsequently put forward, in the form of partial derivatives, which is successfully applied to cause identification when the performance is assessed to be non-optimal. Finally the validity of the proposed approach is illustrated through TE process.