Chiller Performance Evaluation and Optimization Algorithms for Existing Buildings

There is a growing need for energy efficient air-conditioning systems as climate change increases cooling loads and is accelerated by resultant GHG emissions. Residential buildings are a prime candidate for Retro-Commissioning (RCx) as their control algorithms are not often reviewed or revised post construction and commissioning. A data-driven approach is implemented to create regression models predicting chiller power consumption and building thermal response using readily available data in residential Building Automation Systems ("BAS") and weather data. Best case models predict chiller power draw within RMSE 3.1 kW o (0.52% of full load) over the cooling season and building thermal response of RMSE 0.71 C. This research also investigated wrapping models into a model predictive control ("MPC") algorithm using Bayesian Optimization to update chiller setpoint parameters in real-time to reduce chiller power consumption. MPC algorithms deployment is simulated using data logged from two typical multi-unit residential buildings in ASHRAE Climate Zone 5. Best case simulated optimizer performance showed potential chiller energy savings of approximately 12%.