A Belief Rule Base Inference Method For Process Alarm Prognosis With Applications

With the continuous increase of the production capacity and the complexity of the process industry, safety performances are becoming more and more widely concerned. Therefore, the process alarm system is gradually becoming a hot application research issue. Among them, the alarm prediction can make early judgments on abnormal conditions that may occur in the process, which can inform the operators to make corresponding decisions in order to reduce the probability of accident occurrence, and do better protection for the safety of production equipment and operators.In this thesis, a method based on belief rule base inference with time series prediction metrics for process alarm states prognosing is proposed. Firstly, the structure of the belief rule base model and the evidential reasoning algorithm are studied. Regarding unknown parameters involved in the reasoning model, the particle swarm optimization algorithm is used to complete the parameter learning process. The time series historical data are used as the prerequisite attribute and the predicted data are used as the result attribute in order to create the prognosing model. Numerical examples are given to demonstrate the feasibility of this method. Next, for different process variables, single & multi variable process alarm prediction methods are employed to train and verify the prognosing model, resulting in the online application prognosing models. Finally, the proposed method is applied to the DMF recovery process. Belief rule base inference models are established before the prognosis of the process alarm states are achieved successfully. In addition, an intelligent alarm state prognosing system is developed preliminarily. The research results show that the method proposed in this paper can effectively realize the prognosis of the process alarm state.