贝叶斯网络
动态贝叶斯网络
概率逻辑
计算机科学
数据挖掘
条件概率
假警报
风险评估
风险分析(工程)
事件(粒子物理)
警报
残余物
钥匙(锁)
可靠性工程
机器学习
人工智能
工程类
作者
Guoming Chen,Jingyu Zhu,Guoming Chen,Zhigang Tian
标识
DOI:10.1016/j.ress.2022.108700
摘要
• Real-time probabilistic risk assessment for petrochemical industries. • A DBN proposed for incorporating data monitoring in risk analysis. • Both prior risk and data monitoring are considered. • CPTs in DBN are constructed without additional knowledge and accident data. Safety is of high societal concern in the petrochemical industry. With advancing digitalization, the techniques of probabilistic risk assessment (PRA), which provide a system-level perspective for industrial risk analysis, have become increasingly dynamic. This paper provides a further advancement in this direction by proposing a risk updating method based on the dynamic Bayesian network (DBN) to incorporate data monitoring into PRA in real-time. We update the probabilistic risk for basic events in Bayesian networks based on their prior probability and the probability of their online monitoring signals exceeding the alarm threshold. The key idea behind this approach is that if the signal of a basic event exceeds the alarm threshold, the corresponding risk should increase. Otherwise, the basic event should have a low risk. The contribution in this paper is twofold. First, residual methods are developed to estimate signal probabilities exceeding abnormal thresholds. Second, a DBN model is proposed to integrate prior risk with data monitoring for risk analysis. The proposed DBN model does not require additional expert knowledge and historical accident data to define the conditional relationship between data monitoring and prior risk. The proposed approach is validated using the RT 580 experimental setup and managed pressure drilling operations.
科研通智能强力驱动
Strongly Powered by AbleSci AI