Thermal Power Plant Process Monitoring using Mutual Information and Distributed Statistical Model

In this paper, a distributed modeling method is proposed to monitor a practical thermal power plant, which consists of a large number of process variables as well as huge number of data records. In order to simplify the monitoring task, the process variables are decomposed into several sub-blocks by investigating the mutual information matrix estimated from data. Based on the data samples in each sub-block, a set of monitoring models are constructed using principal component analysis(PCA). The monitoring statistics are then integrated using Bayesian method. Whenever a fault is observed, the sub-blocks that are affected by the fault are analyzed and faulty variables are isolated using contribution plot. Finally, the fault root causes are diagnosed using causality analysis. Through the analysis of process data from a practical power plant, the feasibility and efficiency of the proposed method is verified.