氧化铝生产的蒸发过程故障检测仿真研究

Fault Detection Simulation for Evaporation Process of Alumina Production

研究氧化铝蒸发过程故障检测问题。氧化铝蒸发过程是一个非线性,时变的过程,结构错综复杂,样本数据少和难以建立精确模型进行实时检测问题。为此,提出了一种采用贝叶斯网络的小数据集蒸发过程故障检测方法。首先,在贝叶斯理论基础上,提出了一种结合领域知识的贪婪结构学习算法,由领域知识引导加边、减边和转边算子搜索,加速找到评分最高的网络结构。然后,小数据集通过Bootstrap抽样法获得大样本数据,经由改进的结构学习算法得到贝叶斯故障检测网络,检测网络包含变量的拓扑结构图及其条件概率表。用已知数据对检测网络进行验证,结果表明上述网络是有效的,同时亦能锁定故障所在位置,为蒸发过程实时优化控制开拓了一种新思路。

A small datasets fault detection method for Alumina evaporation process based on Bayesian network is proposed. Firstly, a greedy structure learning algorithm combined with domain knowledge is studied. Guided by the domain knowledge, three kinds of searches are conducted ： edge - insertion, edge - deletion and edge - reversal, so as to find the highest score of the network structure fast. After that, a large sample of data is obtained by Bootstrap sampling method from a small datasets and Bayesian fault detection network of Alumina is calculated by the improved structure learning algorithm which includes variable＇s topology structure and the conditional probability tables. Finally, the diagnostic network is tested with the known data. The simulation proves that this method is accuracy and can also lock the location of the fault at the same time, which provides a new idea for the real - time control of the evaporation process.