Enhancing power security: Attack detection based on cost-sensitive SVM and attention mechanism

In order to proactively block attack paths and secure the power monitoring host, a cost-sensitive SVM-based method for detecting attacks on power monitoring hosts is proposed. This method aims to accurately identify network and user attacks. A model for detecting power monitoring host attack behavior using a cost-sensitive SVM approach is developed. The data on power monitoring host attack behavior, encompassing both user and network attack behaviors, is gathered, and the data characteristics are used to create training samples for the model. By resolving the misclassification cost of each sample using a dynamic cost function based on class membership, the cost-sensitive SVM classifier is trained to detect attacks. The classifier then processes the data for classification and other operations to output the detected instances of attack behavior on the power monitoring host. Experimental results demonstrate that this approach can effectively reduce the overall misclassification cost of power monitoring host attacks. It can detect network attacks and user attacks by analyzing abnormal network traffic and user keystrokes, as well as accurately identify other types of attacks on power monitoring hosts. This method enhances the detection capabilities of power monitoring host attacks, thereby ensuring the security of power systems.