Conference Year
July 2018
Keywords
Optimization, control, extremum seeking, chiller
Abstract
In the last 30 years, research into energy optimization via supervisory control of building chiller plants has yielded promising approaches ranging from detailed physics model based optimization to data based or model free optimization. In general, while physics based models tend to require domain expertise and significant calibration effort for implementation, data based and model free approaches require less calibration effort but may be risky to implement due to concerns about robustness to normal variations in environmental conditions and changes to the number of chillers in operation. This paper focuses on rigorous experimental evaluation and verification of model free extremum seeking control, a real-time gradient descent optimization tool. There have been several publications illustrating the effectiveness of extremum seeking control applied to a variety of heating, ventilation, and air conditioning plants in simulation and on mini-split ductless air conditioning system test beds. However, possibly due to inaccessibility of commercially operational chiller plants for experimentation, an evaluation of extremum seeking has not been documented for a large scale in-service building chiller plant. In this paper, a hybrid extremum seeking and model based supervisory control approach is applied to a 2000RT commercial building chiller plant at Chinatown Point mall in Singapore. The extremum seeking control algorithm selects a set point for the condenser water pump flow rate in order to find the value that minimizes the chiller plant's energy consumption. Meanwhile, a model of pump power consumption is used to select the number of operational pumps in response to the ESC's commanded flow rate in order to minimize pumping power. The extremum seeking controller is initialized using settling time and curvature parameters obtained from a several hour long field identification experiment, while data from a longer multi-day experiment provides a control performance verification model. Evaluation and verification of the controller took place over a testing period lasting 1 week and cycling through morning, daytime, evening, and night modes of operation. Comparing the experimental results against the data based control verification model’s predictions confirms that the extremum seeking approach is able to find the globally minimizing condenser water pump flow rate over a normal range of environmental conditions and chiller plant configurations. The paper concludes with recommendations for how the approach can be applied to plants of similar size but different architecture and how inputs can be added to the extremum seeking control algorithm to get the plant closer to its optimal settings.