Industrial control systems play a crucial role in society, and ensuring the security of industrial control systems has become an important and challenging task. Intrusion detection is an important defense mechanism for protecting these systems. However, in existing industrial control intrusion detection techniques, especially those based on deep learning, a large number of training samples are usually required, and real-world samples often exhibit data imbalance, which can affect the training effectiveness. In this paper, a network intrusion detection model based on domain adaptation is proposed, which is a sample-discriminating adversarial domain adaptation model that can classify all samples of the target domain in a universal domain adaptation scenario.