Intelligent Attack Vector Classification for Smart Grid CPS

The application of artificial intelligence in the security of smart grid cyber-physical systems (CPSs) faces challenges arising from the "black box" characteristics of standard deep learning models, which lack opacity on the safety side. There is an important task of developing an accurate-security framework that will be transparent. The proposed research will focus on this important issue. The research proposes an innovative "Neural Network Forest (NNF)," which will focus on an "explainable by design" approach concerning anomaly detection and resilient controls. The proposed innovative "NNF" will be an "expert" ensemble of various "NNs" on specific sub-domains of the grid. The outputs of all "NNs" will then be combined by an "Attentive Aggregation Mechanism," which will provide not only an "aggregate system classification" outcome but also an interpretable "diagnosis of the location of the anomaly." For resilient controls, the "NNF" will use its transparent outputs on a "decoupled policy mapper." The proposed "NNF" will be evaluated on the "IEEE 39-Bus System." The proposed "NNF" was able to provide "99.3% accuracy" in "attack detection," which was on par with standard models. However, it was able to provide an additional "95.2% Diagnostic Precision" concerning "locating the sources of attacks." Importantly, this measure could not be evaluated in standard "black box models." The "significance" of this research will add an "important contributory chapter in the development of trusted Al on CPSs" by making smart grid "resistance more realistic with increased confidence in smart grid intelligence. Keywords— AI, Deep Learning, Reinforcement Learning, Neural Network Forest, Smart Grid, cybersecurity, Cyber–Physical Systems