Combining Precursor and Cloud Leaky Noisy-or Logic Gate Bayesian Network for Dynamic Probability Analysis of Major Accidents in the Oil Depots

Major accidents in oil depots are low-frequency/high-consequence events. Because of the relative scarcity of accident data, it is difficult to elucidate the dynamic characteristics of risks using conventional methods. Direct data on major accidents is scarce. Thus, relevant data on precursor accidents has attracted increased attention. Here, the Cloud Leaky Noisy-OR(CLNOR) logic gate is proposed to improve the traditional Bayesian network (BN), and a probabilistic analysis model is developed for the analysis of major accidents based on precursor data and Hierarchical Bayesian Analysis (HBA). The CLNOR logic gates extensively reduce the evaluation workload of the traditional noise-OR logic gate. Furthermore, the proposed approach overcomes the cognitive uncertainty introduced by expert elicitation. HBA based on precursor data extracts the dynamic character of risk and deals with the source-source uncertainty introduced by different data sources, thus improving the precision of frequency estimation. The BN allows the dynamic analysis of probabilities and dynamic mining of key risk prevention factors, overcoming the model uncertainty of traditional models. As updates based on new observations are performed, dynamic risk probability distributions are generated. A case study based on the proposed method was conducted, demonstrating that the method is effective for dynamic risk prediction and prevention.