Abstract:

In order to solve the problem of strong subjectivity of the prior data of the dynamic Bayesian network (DBN) obtained when analyzing the reliability of the system, BP neural network was used to optimize the prior data of DBN, taking the burner system of gasifier as the research object. According to the empirical formula of the number of neurons in the hidden layer, the DBN of the gasifier burner system was divided into three subsystems, which were transformed into BP neural network respectively, and the estimated prior distribution of DBN was corresponding to the input function and output function of the BP neural network respectively, the performance of the system is studied and the DBN parameters are optimized by using the characteristics of information is transmitted forward and error backward of BP neural network. The two?way reasoning of the gasifier burner system DBN is carried out to realize the dynamic reliability analysis of the gasifier burner system. The results show that the forward reasoning of the gasifier burner system DBN can obtain the optimized system reliability trend; the reverse reasoning is performed to obtain that the results of key events and weak links remain the same before or after optimization and the weak links are the high value and fluctuation of oxygen coal ratio.

Key words: Bayesian estimation, Monte Carlo simulation, BP neural network, Dynamic Bayesian network, Reliability

摘要：

为解决在分析系统可靠性时获取的动态贝叶斯网络（DBN）的先验数据主观性强的问题，以气化炉烧嘴系统为研究对象，利用BP神经网络优化DBN的先验数据。依据隐含层神经元数量经验公式，将气化炉烧嘴系统DBN模型划分为3个子系统，并分别转化为BP神经网络。将DBN的先验分布分别对应BP神经网络的输入函数与输出函数，再利用BP神经网络信息向前传、误差向后传的特性，对系统进行性能学习，实现对DBN的先验数据优化。对优化后气化炉烧嘴系统的DBN进行双向推理，实现对气化炉烧嘴系统动态可靠性分析。结果表明，对气化炉烧嘴系统DBN进行正向推理，可得到优化后的系统可靠性变化趋势；进行反向推理，可得到优化前后的关键事件及薄弱环节，其中薄弱环节为高氧煤比氧煤比的波动。

关键词: 贝叶斯估计, 蒙特卡洛模拟, BP神经网络, 动态贝叶斯, 可靠性