Berkeley Lab is testing a model predictive control, a control strategy that makes use of a system model, disturbance forecasts, system operational data, and an optimization algorithm to forecast future system behavior over a relevant control horizon and optimize control actions to meet an optimization objective and within specified constraints. This technology targets building-level centralized cooling and heating plants with or without active thermal energy storage. The technology will be demonstrated to control a commercial building office Heating Ventilation and Air Conditioning (HVAC) system that is composed of four variable air volume rooftop units in response to dynamic grid signals.