Monitoring and early warning method for abnormal conditions in complex processes based on bidirectional causal reasoning and its application in diesel hydrotreating plants

Under conditions of high temperature, high pressure, and other extreme factors, petrochemical plants are highly susceptible to abnormal conditions and accidents, which are difficult to trace the causes and predict the consequences. To address these issues, a monitoring and early warning method for abnormal conditions in complex processes based on bidirectional causal reasoning is proposed. Firstly, bidirectional causal reasoning models for abnormal conditions in complex processes are established. This involves constructing a risk identification and analysis knowledge base derived from HAZOP analysis reports. Data from the knowledge base are processed and trained models using algorithms such as LDA clustering, Apriori association analysis, and Naïve Bayes classification. These processes yield categorized causes and consequences, as well as causal coupling relationships, culminating in the establishment of causal clustering models and the risk prediction model. Subsequently, the sensing data are integrated into the Distributed Control System (DCS), Mechanical Condition Monitoring System (CMS), and Environmental Monitoring System according to the different monitoring entities under complex operating conditions. The typical parameters, critical monitoring locations, and risk points for each category of sensing data are analyzed in detail. Based on the monitoring units, parameters, and abnormality descriptions involved in the sensing data, a single abnormal condition text is generated in real-time. By combining the temporal sequence of abnormal conditions with the physical spatial order of abnormal entities, causal relationships among individual abnormal conditions are established, enabling the construction of a real-time and comprehensive description of complex operating conditions. Then, natural language segmentation algorithms and causal clustering models are employed to compute the posterior probabilities of the comprehensive description of complex operating conditions belonging to each category of causes and consequences in the risk identification and analysis knowledge base. Corresponding relationships between the comprehensive description and the classifications of causes and consequences are established. Based on the identified causal coupling relationships, potential causes leading to the current abnormal condition and the possible resulting consequences are determined. Finally, the Naïve Bayes risk prediction model is utilized to perform a risk assessment of the accident scenarios derived from the causal analysis of abnormal conditions. The basic information, comprehensive description, causes, consequences, and risk information of the abnormal conditions are stored in the risk identification and analysis knowledge base, thereby completing the monitoring and early warning of abnormal conditions and bidirectional causal reasoning. This paper presents an application example of the method in a diesel hydrotreating plant, demonstrating that the research provides a methodological and technical basis for the monitoring and early warning and emergency response of abnormal conditions.