Experimental validation of mixed-integer Model Predictive Control for energy management in an industrial food processing plant

This paper presents the development and implementation of a broadly applicable Energy Management System (EMS) based on model predictive control (MPC) to optimize energy consumption in a real-world industrial food processing plant. The EMS, formulated as a Mixed-Integer Linear Programming (MILP) optimization problem, is designed to minimize energy use and switching operations- defined as the number of equipment on/off transitions per unit of energy delivered (switches/MWh) - while ensuring sufficient heating and cooling for production. The control structure is built upon a two-tiered MPC framework. At the higher level, an MPC algorithm optimizes energy efficiency over a 24-hour horizon, taking into account the production schedule, predicted energy demands, and the operation of thermal storage and heat pumps. The lower-level controller, with a faster sampling rate, focuses on short-term disturbance rejection and immediate system adjustments. The system was evaluated over 14 days of real-world economic plant operation, with results showing significant improvements in efficiency and in reducing switching operations and thus wear. On the cold process side, switching operations have been reduced while maximizing control performance under tight temperature constraints. On the hot side, the EMS achieved a remarkable 8 % increase in efficiency and 36 % reduction of switching operations. • Successful real-world deployment of MILP-based MPC in food-industry energy management. • Two-layer predictive control ensures robust, cost-efficient thermal system operation. • Model-based EMS reduces switching by 36% and improves hot-side efficiency by 8% • Predictive load adaptation leverages production schedules to minimize energy use. • Control strategy is transferable to other sectors with batch processes and thermal loads