The Framework of Invariant Electric Vehicle Charging Network for Anomaly Detection
Yu-Wei Chung, Mervin Mathew, Cole Rodgers, Bin Wang, Behnam Khaki, Chi‐Cheng Chu, Rajit Gadh
Abstract
Electric vehicle (EV) charging management systems control and schedule EV load according to the measurements of local building load, solar generation, and dynamic electricity price. Within this information network, any data replaced or modified by an attacker will disrupt the EV charging schedule and could cause damage to the electricity grid. Under real circumstances, these measurements are correlated in a way that is not true for false data. This paper examines the relationship of pairwise measures within the system to establish a correlation-invariant network, and a multivariate time-series segmentation method along with a weighted k nearest neighbor (kNN) classifier is proposed to detect the changes in correlations and identify anomalous data within the network.