Dynamic Graph-Based Anomaly Detection in the Electrical Grid

Given sensor readings over time from a power grid, how can we accurately detect when an anomaly occurs? A key part of achieving this goal is to use the network grid sensors quickly detect, in real-time, any unusual events, whether natural faults or malicious, occur on grid. Existing bad-data detectors industry lack sophistication robustly broad types anomalies, especially those due emerging cyber-attacks, since they operate single measurement snapshot at time. New ML methods are more widely applicable, but generally do not consider impact topology change measurements and thus cannot accommodate regular adjustments historical data. Hence, propose DynWatch , domain knowledge based topology-aware algorithm for detection using placed dynamic Our approach accurate, outperforming existing approaches by 20 $\%$ (F-measure) experiments; fast, averaging less than 1.7 ms per tick 60K+ branch case laptop computer, scaling linearly with size graph.