A Fault Diagnosis Method Based on Canonical Correlation Analysis and Transfer Learning in Wastewater Treatment Plant

Methods of multivariate statistics have gained remarkable popularity in the last few decades. However, the lack of data in the target domain often limits the performance of fault diagnosis in industrial processes. To solve this problem, this paper proposes a fault diagnosis approach based on canonical correlation analysis and transfer learning. Different from the standard method, this method has the following characteristics. Firstly, based on the canonical correlation analysis method, the diagnosis model of source domain can be obtained using a large amount of source domain data. Secondly, the transfer learning method based on the local retention canonical correlation analysis of K nearest neighbors transfers the fault diagnosis model of the source domain to the target domain, which alleviates the problem of insufficient data in the target domain through migration. Thirdly, a process monitoring statistic based on the Riemannian distance is proposed. Finally, the data from the wastewater treatment plant are analyzed to demonstrate the effectiveness of the proposed method.