TO-22-C019 - Real-Time Implementation of Model Predictive Control for Cooling System of a Factory Building

This paper presents a real-time application of model predictive control (MPC) for a cooling system of a target building using machine learning models. The building isa heavy-machinery assembly building consisting of many large indoor spaces. The size of a thermal zone of our interest is 80 m*60m*9.7m(262ft*197ft*32ft). The zone is served by four condensing units (210kW(59.7RT) x 4EA), two direct expansion type AHUs (40,000m3/h(1,412,588ft3/h) x 2EA) and 45 diffusers. To implement the MPC for the thermal zone, the authors developed two simulation models using Artificial NeuralNetwork (ANN): one is to predict future indoor air temperatures at multiple indoor points with the inputs of current supply air temperature, outdoor airtemperature, indoor air temperature, the status of condensing unit operation (on/off) and status of fan operation(on/off). The other is to predict future supply airtemperature from AHUs with the inputs of current outdoor air temperature, the status of condensing unit operation (on/off) and status of fan operation, and outdoorair's relative humidity. The accuracy of the two models was assessed in terms of MBE and CVRMSE (MBE=1.2%, CVRMSE=3.8%). The objectivefunction of MPC is to minimize the energy use of four condensing units while maintaining indoor temperature lower than or equal to the indoor setpoint temperature.The proposed MPC was realized in the target building at the sampling time of 10 minutes for four weeks in August-September 2021. It was found that theproposed control saved energy by 35.1%. Finally, the authors outline issues that could occur while performing the MPC of a factory building, as well as thepotential of the MPC for indoor environmental control and energy savings.