Application of Machine Learning Algorithms in the Optimal Design and Operation of a Chiller Plant with Thermal Energy Storage

Abstract

Several studies show that Heating, Ventilation, and Air Conditioning (HVAC) systems consume more than 40% of the total building energy. In order to reduce building energy consumption, it is necessary to optimize HVAC system energy consumption through proper design and operational strategy. The present study aims to do that by retrofitting a thermal energy storage (TES) system with an existing chiller plant located in an academic complex. In order to arrive at optimal design of the stratified chilled water TES, it is proposed to use a machine learning algorithm, using which the building cooling load can be predicted accurately based on the available plant data. The optimal design of TES involves analysis of various operating schedules based on the actual building cooling load and air conditioning plant capacity. By decision tree regression of power consumed with respect to load as percentage of chiller capacity, it was found that the existing chiller was operating at low load with low COP, also by K-Means clustering analysis these findings were proved right. Polynomial linear regression was used to find optimum chiller utilization capacity which in this case was at 56.4%. Results show that addition of TES can reduce the chiller energy consumption by as much as 14.2%. If one considers the Time-of-Day tariff, then the savings in energy cost can be much higher.