Research on Network Intrusion Detection of Automation System Based on Machine Learning

In the context of the increasingly urgent demand for network security protection of automation systems, this paper proposes a new network intrusion detection algorithm that integrates attention mechanism and spatiotemporal feature mining. The algorithm constructs a hybrid neural network structure based on Transformer and LSTM, and accurately captures complex intrusion features in network traffic through an adaptive weight adjustment strategy. The experiment uses KDD Cup 99 and CICIDS 2017 datasets, with accuracy, recall, F1 value and false alarm rate as evaluation indicators. The results show that on the KDD Cup 99 dataset, the algorithm has an accuracy of 98.7%, a recall of 97.5%, an $F 1$ value of 98.1%, and a false alarm rate of 1.2%; on the CICIDS 2017 dataset, the accuracy is 97.3%, the recall is 96.8%, the F 1 value is 97.0%, and the false alarm rate is 1.8%. Compared with traditional detection methods, this algorithm shows stronger adaptability in identifying new attacks and variant attacks, significantly reducing the risk of missed reports and false positives, and providing an efficient and reliable solution for network intrusion detection in automation systems.