Anomaly Detection of Power Information System Based on Attention Mechanism CNN-LSTM

The power information system is in charge of keeping track of each power grid platform's condition and making sure that each platform runs reliably. Power information system log anomaly detection has numerous challenges, including a vast amount of data, a high level of attack threat concealment, significant fault damage, and sophisticated feature engineering of conventional methods. Based on the timing of log data. An attention-based CNN-LSTM anomaly data detection model of the power information system is developed based on the timing of the log data. Learning algorithm with neural network training is used as the model's input, and CNN is then used to extract the features of the data in order to model the logging theme pattern like a speech recognition sequencing. Next, a temporal form of the feature vector is fed into the LSTM. Finally, the network parameters are optimized using the attention approach to achieve log anomaly detection. The suggested model outperforms CNN and CNN-LSTM in terms of detection performance, achieving an average log anomaly detection accuracy of 93%, according to analysis and experimental data.