Lawrence Berkeley National Laboratory

To implement the condenser water set point optimization, one can employ a regression model. However, existing regression-based methods have difficulties to handle non-linear chiller plant behaviour. To address this problem, we develop a Bayesian network model and compare it to both a linear and a polynomial regression model via a case study. The results show that the Bayesian network model can predict the optimal condenser water set points with a lower root mean square deviation for both a mild month and a summer month than the linear and the polynomial models. The energy-saving ratios by the Bayesian network model are 25.92% and 1.39% for the mild month and the summer month, respectively. As a comparison, the energy-saving ratios by the linear and the polynomial models are less than 19.00% for the mild month and even lead to more energy consumption in the summer month (up to 3.73%).