Data-Driven Based FDIAs Detection and Sensitive Feature Identification for Cyberattack Defending of Renewable Energy.

With the continuous increase of renewable energy integration into the power grid, cyber-physical systems play an increasingly significant role in the smart grid. Most renewable energies connect to the power system with grid-connected converters (GCCs). However, the control of the GCCs unavoidably relied on the communication-based signal, which increases pressure on data interaction security in cyber-physical power systems (CPPS). To address this issue, in this paper, a novel importance index and its defining method based on the F1-score and probability of false data injection attacks (FDIAs) are proposed to filter the sensitive feature. In the proposed defining method, data preprocessing is achieved by local outlier factor (LOF) based on k-maxmin clustering algorithm and the imputation method based on random forest (RF). The principle component analysis (PCA) reduces the dimensionality of multi-dimensional features and further filters the redundant information. Based on the defining method, a methodology based on convolutional neural networks and support vector machine (CNN-SVM) is proposed to detect the types of cyberattacks. Different from the traditional CNN classifier, the softmax is replaced by SVM, which maps various features to different hyperplanes to achieve more accurate classification. To analyze the types of features, the sensitive feature is defined according to an importance index and used to strengthen cyber defenses by focusing on monitoring such feature data. In order to verify the effectiveness and feasibility of the proposed method, experiments based on real data from the eastern province of China are adopted. The experiment results indicate that the proposed method could find the weaknesses of the dataset, which are immunity to cyberattacks.