Real-time updating strategy for Bayesian network-based coal mill process abnormity diagnosis model

Coal mill is the core equipment of coal pulverizing system in the thermal power plant. It is of great significance for system safety to formulate the abnormity diagnosis model based on a small amount of data when the new coal mill is into production. In this paper, the diagnosis model based on three typical abnormities in the process of coal mill is firstly established. A new real-time updating strategy for the Bayesian network (BN) model based on node identification is proposed. Taking the abnormity diagnosis BN of existing coal mills as the source domain model and using the small amount of new data of the coal mill in the target domain, the nodes that do not match the new information could be found out. Retaining the useful information of the source domain model, the target domain model will be updated and supplemented according to the new data through local updating. To verify the proposed method, the method is applied to the diagnosis process of abnormity. Experimental results show that the updated model has good performance, and the average correct rate of diagnosis is more than 98%. © 2021, Science Press. All right reserved.