Optimizing DER dispatch in a renewables dominant distribution network using a virtual power plant

An optimal dispatch controller was used to optimize net load of a virtual power plant consisting of MW-scale solar PV, battery energy storage, and flexible customer loads on a solar-dominant distribution network in Shirley, MA. A regression-based prediction methodology was used to predict net load over a rolling 24-hour time horizon, with demonstrated accuracy of 8 to 12% root mean square error; net load predictions were ingested by a simulated-annealing algorithm to generate real-time optimal dispatch schedules for system assets given user-defined objectives. Multiple objective functions were tested in combination, including peak shaving, energy cost optimization, peak-power dispatch, and power firming. Field test results show that integrating time-series load and solar predictions and flexible loads into dispatch decisions effectively increased storage capacity relative to a non-predictive baseline by upwards of 20%. Results further highlight the need to manage prediction uncertainty, for example by maintaining sufficient BESS reserve capacity.