Design of Intrusion Detection and Response Mechanism for Power Grid SCADA Based on Improved LSTM and FNN

The current behavior pattern extraction methods in intrusion detection systems cannot fully extract information. To improve the accuracy of such systems, the study first uses sequence feature construction algorithms to explicitly represent sequence feature information. Afterwards, an intrusion detection system is designed that combines long short-term memory networks and feed-forward neural networks to remember sequence information and adjust output dimensions, thereby mapping the results to classification labels. According to the simulation comparison results, the designed system had a significantly higher packet capture rate per second compared with the other three intrusion detection systems. When the intrusion rates were 10% and 22%, respectively, the designed system had a packet capture rate of 7000ps per second and a system occupancy rate of 23%. Compared with other intrusion detection systems, the intrusion detection and response mechanism of the proposed power grid monitoring and data acquisition control system was more outstanding in terms of functionality and practicality. The loss value of the proposed method was less than 0.22, and the F1 value was 99.97%. The F1 value of the model combining convolutional neural network and bidirectional long short-term memory network improved by 0.90%. This indicates that the proposed method has higher accuracy and reliability in intrusion detection tasks. This study contributes to the development of power system security protection technology, providing important references for intrusion detection research in other fields.