基于潜空间协整自编码的化工过程特征提取与监测OA

Modern chemical processes often exhibit both significant nonlinearity and dynamic non-stationarity due to their complex internal mechanisms.Traditional single model approaches struggle to fully characterize these properties for effective process monitoring.To address this challenge,a process monitoring method that integrates convolutional autoencoder(CAE)and cointegration analysis(CA)was proposed.The method first utilized the nonlinear feature extracting capability of CAE to map high-dimensional raw data into a low-dimensional latent space,which represented the core process dynamics.Subsequently,CA was applied to these latent features to capture the long-term equilibrium relationships among variables.Finally,the feasibility of the method was verified by real industrial production data.The results showed that the proposed method achieved a fault detection rate(FDR)of 99.53％with a false alarm rate(FAR)of only 0.55％.Its overall performance significantly surpassed that of standalone methods such as CAE or CA,which also proved that the method could provide an effective solution with high sensitivity and high reliability for the safety monitoring of complex industrial processes.