Abstract: In view of the multistage, strong coupling, and nonlinearity of batch process data, we propose a fault diagnosis method for the batch process based on kernel entropy component analysis (KECA), improved gray wolf optimization, and the kernel extreme learning machine (KELM). Considering the multistage nature of batch process data, the K-means algorithm is used to cluster these data, and the whole process period is divided into several sub-periods. Based on the strong coupling and nonlinearity of batch process data, we introduce the KECA algorithm to extract the features of the original fault data and obtain their deep features. Using KELM as a classifier and the improved gray wolf algorithm, which improves the initialization strategy and convergence factor, the parameters of the classifier are intelligently optimized and the optimal classifier is obtained by diagnosing the faults of each period of the batch process. We verified the feasibility and superiority of this method based on the results of a simulation experiment and a contrast experiment with penicillin.