With the integration of industrial control systems (ICSs) and modern IT networks, the security of ICSs has been threatened while increasing their efficiency. Existing intrusion detection methods based on machine learning, such as Support Vector Machine (SVM), Decision Tree, etc., usually rely on manually designed methods for feature learning and have low accuracy for intrusion detection of high-dimensional network traffic of ICSs. Although the detection accuracy of Long Short-Term Memory (LSTM) and Gated Recur-rent Units (GRU) based methods is significantly improved compared to Simple Recurrent Neural Network (SimpleRNN), there is the problem of long training time consumption. To solve the above problem, this study proposed an intrusion detection method for ICSs based on 1D Convolution Neural Networks (1D CNN) and Bidirectional Simple Recurrent Unit (BiSRU), fully learning the correlation and dependency of network traffic data of ICSs in spatial and temporal dimensions. With skip connections employed, the optimized bidirec-tional structure of the Simple Recurrent Unit (SRU) neural network can further alleviate the problem of gradient vanishing and improve the training effect. Mississippi State Uni-versity's Gas Pipeline dataset was used to train and test the model. Experiments show that the proposed method is significantly better than other existing methods in terms of accu-racy and training time.