基于核熵成分分析结合独立元分析的故障检测方法

摘要

Aiming at the complex industrial process with strong nonlinearity, non-Gaussian and many variables, a fault de-tection method based on kernel entropy component analysis ( KECA) and independent component analysis ( ICA) was proposed. In the method, the data dimension was reduced by KECA, ensuring the minimum information loss.Then the score matrix was re-solved by ICA;and according to the I2 and SPE statistic, the fault can be found.The simulation results on Tennessee Eastman process illustrate that the method is full of feasibility and effectiveness.Moreover, the robustness of detection results is analyzed.%针对工业过程具有多变量、非线性、非高斯等特点，提出了一种基于核熵成分分析与独立元分析的（ KECA－ICA）的故障检测方法。首先通过核熵成分分析对数据进行降维，保证了信息量损失最小；然后对熵成分的得分矩阵进行ICA分解，并根据监测量SPE和I2的状态判断系统是否发生故障。通过对TE（ Tennessee Eastman）过程的仿真研究，验证了该方法的可行性与有效性，并且对检测效果的鲁棒性能进行了分析。