Anomaly Detection in the Distribution Grid: A Nonparametric Approach

Owing to the development of smart grids technologies, including renewable energy resources and integrated demand response, the traditional distribution network undergoes systemic improvements and has become an active system. To meet such changes in technology, the smart distribution grid is highly dependent on communication infrastructures. Although this strong reliance can make the distribution network susceptible to cyber threats, there has been inadequate attention to the distribution network cyber-security. An intruder can disrupt communication and effectively target the distribution management system core functions, leaving the distribution grid susceptible to loss of stability boundaries. One means of ensuring the stability of the distribution system is to explicitly utilize proven algorithms for transmission grid's attack detection. Nevertheless, various features of the distribution grid may trigger current algorithms to malfunction. This paper attempts therefore to explain these unique features and to integrate them into the attack detection algorithm. For example, for designing detection mechanism, we utilize the high r/x ratio, tree structure and dynamic variation of load profiles. Afterwards, using a kernel density estimation technology together with statistical calculations, an unsupervised anomaly detection procedure is suggested. This follows by an outlier detection to identify attacks. We validate the proposed method on the IEEE 33-bus test system. Simulation results demonstrate that the proposed algorithm correctly identifies the distribution system anomalies at a high detection rate.