An Intelligent Approach to Develop, Assess and Optimize Energy Consumption Models for Air-Cooled Chillers using Machine Learning Algorithms

Abstract

The building sector accounts for more than 70% of thetotal electricity use. Chillers consume more than 50% of electrical energyduring seasonal periods of building use. With the growth of the building sectorand climate change, it's essential to develop energy-efficient HVAC systemsthat optimize the ever-increasing energy demand. This study aims to develop anenergy consumption prediction model for air-cooled chillers using machinelearning algorithms. This is done by developing different static and dynamicdata-driven regressive and neural network models and comparing the accuracy oftheir prediction to identify the most accurate modeling algorithm using 3 maininputs chilled water return temperature, outside drybulb temperature, andcooling load. The proposed model structure was then optimized in terms of thenumber of neurons, epochs, time delays aswell as the number of input variables using a genetic algorithm. Training andtesting were done using real data obtained from a fully instrumented 4-tonair-cooled chiller. Results of the study show that the optimized artificialneural network model can predict energy consumption with a high level ofaccuracy compared to conventional modeling techniques. The development ofhighly accurate self-tuning models can be a powerful tool to use for otherapplications such as fault detection and diagnosis, assessment, and systemoptimization. Further studies are necessary to evaluate the effectiveness ofusing deep learning algorithms with more hidden layers and cross-validationtechniques.