基于高斯受限玻尔兹曼机的非线性过程故障检测

针对非线性工业过程,提出了一种基于高斯受限玻尔兹曼机(GRBM)模型的故障检测方法.该方法从海量过程数据中提取出GRBM隐层特征信息,通过隐层特征再构建出重构数据,并依据重构误差在残差空间中构建检测统计量,形成了非线性过程故障检测算法.仿真结果表明,基于GRBM的故障检测方法不仅比传统的核主元分析(KPCA)方法具有更好的故障检出率,并且针对大数据量问题具有更强的处理能力.

Nonlinear Process Fault Detection Based on Gaussian Restricted Boltzmann Machine

To monitor nonlinear process, a novel fault detection method based on Gaussian restricted Boltzmann machine (GRBM) model was proposed. Useful features in the hidden layer of GRBM were extracted from original process data and reconstructs data with the features in the visible layers of GRBM. Monitoring statistic was constructed in residual space from reconstruction errors. The proposed algorithm was implemented in monitoring a numerical nonlinear process. Simulation results showed that GRBM method worked better than traditional kernel principal component analysis (KPCA ) method in fault detection rate and dealing with mass data.