基于多块MICA-PCA的全流程过程监控方法

Plant-wide process monitoring based on multiblock MICA-PCA

多块策略广泛应用于全流程过程监控领域,以解决变量关系复杂性较高的问题,但传统分块方法得到的子块数据存在高斯与非高斯混合分布问题,影响过程监控的效果.为此,提出一种基于多块MICA-PCA的过程监控方法.首先采用Jarque-Bera(J-B)检测方法对原始数据进行高斯与非高斯分块;然后利用Hellinger距离(HD)方法获得高斯与非高斯子块,通过对高斯与非高斯子块采用不同的建模和诊断方法,提高监控效果;最后将该方法应用于田纳西-伊斯曼(TE)过程的监控中,以验证所提出方法的有效性.

The multiblock strategy is widely used for plant-wide process monitoring, to capture the relations between complex process variables of the plant-wide process, however, the sub-block data obtained from the traditional multiblock method still has the problem of non-Gaussian and Gaussian mixture distribution, which influences the performance of process monitoring. Therefore, a plant-wide process monitoring method based on multiblock MICA-PCA is proposed to improve the process monitoring performance. Firstly, the measured variables are automatically divided into non-Gaussian block and Gaussian block through the Jarque-Bera（J-B） test method, the non-Gaussian block and Gaussian block are divided into non-Gaussian sub-blocks and Gaussian sub-blocks through the Hellinger Distance（HD） method. By using different modeling and diagnosis methods for non-Gaussian sub-blocks and Gaussian sub-blocks, the monitoring effect is improved. Finally, the proposed method is applied to monitor the Tennessee-Eastman（TE） process, which shows its effectiveness.