Causal Model Reconstruction of Complex Energy System Using Probabilistic Network Approach

Abstract

This paper addresses the challenges in performance monitoring and degradation diagnosing of complex industrial process. A Bayesian network-based model was developed using real operation data of the system, and readings from multiple sensor sources were regarded as evidences to identify the root cause of process variations causing system degradation. An inferring process was executed with the belief network to output possible root causes of the process variation. Then, a probabilistic confidence level of controllable variables was recommended for optimizing operation. A full-scope simulator for the power plant unit is employed as an indispensable test bed for the probabilistic network model. To achieve a decision support simulation, a simulated operation database is generated through varying boundary parameters. The model was re-parameterized with the data from the simulated operation database. The simulation test results show that the model can properly represent the domain knowledge, and the decision advices are effective for such an elaborate work. The probabilistic model can tackle the problems which are indiscernible in conventional thermodynamic model-based analysis.