Fast Inverter Control by Learning the OPF Mapping Using Sensitivity-Informed Gaussian Processes

Fast inverter control is a desideratum towards the smoother integration of renewables. Adjusting injection setpoints for distributed energy resources can be an effective grid mechanism. However, finding such optimally requires solving optimal power flow (OPF), which computationally taxing in real time. Previous works have proposed learning mapping from conditions to OPF minimizers using Gaussian processes (GPs). This GP-OPF model predicts when presented with new instance conditions. Training enjoys closed-form expressions, and predictions come confidence intervals. To improve upon data efficiency, we uniquely incorporate sensitivities (partial derivatives) into GP-OPF. expedites process generating training dataset as fewer instances need solved attain same accuracy. further reduce computational approximate kernel function leveraging concept random features, neatly extended sensitivity data. We perform analysis second-order cone program (SOCP) relaxation OPF, whose computed by merely system linear equations. Extensive numerical tests real-world on IEEE 13- 123-bus feeders corroborate merits