Smart Grid Load Balancing with Renewable Forecasting

The Challenge

A regional utility operating 2.4 GW of renewable generation capacity (1.8 GW solar, 0.6 GW wind) was curtailing 18% of potential renewable generation due to grid stability concerns. The grid operator could not accurately predict how much renewable energy would be available in the next 24–48 hours, so they maintained excessive spinning reserve from natural gas plants "just in case." This curtailment represented $31M in wasted clean energy annually and slowed the utility's progress toward its decarbonisation target.

Our Approach

We built a generation forecasting and dispatch optimisation system that predicts renewable output 48 hours ahead and optimally schedules battery storage and demand response to absorb variable generation.

Generation Forecasting: Separate LSTM models forecast solar irradiance and wind speed at each generation site, using inputs from multiple weather model ensembles (GFS, ECMWF, NAM), satellite cloud imagery, and on-site SCADA measurements. The solar model accounts for panel soiling, inverter clipping, and temperature derating. Forecast accuracy at the 24-hour horizon is 94% (normalised RMSE of 6%).

Battery Dispatch Optimisation: A linear programming model schedules charge/discharge cycles for the utility's 400 MWh battery fleet. The objective function minimises total system cost (fuel + curtailment penalty + battery degradation) subject to grid frequency constraints, battery state-of-charge limits, and transmission capacity. The optimiser runs every 15 minutes with a rolling 48-hour horizon.

Demand Response: During periods of excess renewable generation, the system triggers demand response events — primarily scheduling industrial loads (water treatment, cold storage, EV charging) to absorb surplus energy. Participating customers receive real-time price signals through an API integration with their energy management systems.

Results