Online Estimation of Geomagnetically Induced Current Effects using Neural Networks

Geomagnetically-induced currents (GIC) are quasi-DC currents that can flow in the power grid due to space weather. These currents can lead to negative effects such as transformer overheating, excessive harmonics, large reactive power loss, and potential blackouts. To improve real-time situational awareness of system operators, some utilities have installed GIC monitors at the neutrals of critical high-voltage transformers. However, there are no online tools to compute the corresponding transient and steady-state effects of the measured GIC on transformers. This work therefore explores the use of several dense neural networks stacked together to provide near real-time computation of the trajectories of current harmonics up to the 15 ^th order and reactive power consumption for various operating conditions. It was shown that the neural network-based online estimator has an average accuracy of 94% for values above 1 A (or 1 Mvar), and the total computation time requirement on a standard CPU is about 0.53 seconds.