Abstract:

Modern industrial processes always have multiple operation modes. Besides, the variable in the single mode often obey complex data distribution which is a mix of Gaussian distribution and non-Gaussian distribution. Considering the problems of both multimode and complex data distribution, a new multimode processes monitoring method called multiple SVDD is proposed based on the local outlier probability algorithm and the support vector data description algorithm. First, given that the differences exist between different modes, the clustering is conducted by employing the differential strategy and the local outlier probability algorithm. Second, the SVDD algorithm is used to build the monitoring model in each single mode. And then, the most suitable model is selected for each testing sample through calculating the outlier probability. Finally, the feasibility and efficiency are proved through the Tennessee Eastman process simulation.

Key words: multimode, complex data distribution, local outlier probability, SVDD, processes monitoring

摘要：

现代工业过程往往具有多个运行模态,并且单一模态中的变量服从高斯与非高斯混合的复杂数据分布。针对多模态与复杂数据分布问题,基于局部离群概率(local outlier probability, LOOP)算法与支持向量数据描述(support vector data description, SVDD)算法,提出了一种名为MSVDD(multiple support vector data description, MSVDD)的多模态过程监控方法。首先,考虑到不同模态之间存在差异,利用差分策略以及局部离群概率算法对多模态数据进行聚类。其次,在每个单一模态下分别建立SVDD模型。然后,通过计算测试样本对每个单一模态的离群概率选择合适的模型进行过程监控。最后,在Tennessee Eastman(TE)平台上进行仿真测试以验证提出方法的可行性与有效性。

关键词: 多模态, 复杂数据分布, 局部离群概率, 支持向量数据描述, 过程监控