Abstract: With the increasing incidence of cyberattacks on industrial control systems, existing intrusion detection models based on deep learning experience difficulty in determining the optimal hyperparameter set; consequently, detection accuracy is low. To solve this problem, we propose an intrusion detection model that combines hyperparameter optimization algorithms and stacked long short-term memory (SLSTM) for industrial control systems. This model can automatically optimize the hyperparameters of the deep learning industrial control intrusion detection model. Thus, it effectively improves detection accuracy. First, we use the borderline synthetic minority oversampling technique to deal with the problem of sample imbalance in the original data. Second, we apply the Bayesian optimization algorithm to search for the optimal hyperparameter group of SLSTM. Lastly, we compare the proposed model with other related algorithms by using an industrial control network standard data set. Results show that our algorithm has the highest applicability in industrial control intrusion detection.