Optimisation of automated chiller plant for increased energy efficiency

Abstract

Chiller systems are used in buildings all over the world and account for a significant portion of a country’s electrical demand. Advancements in technology, as well as improvements in controls, have contributed to a reduction in their impact on consumption. This thesis is aimed at developing and testing an algorithm for the condenser side of a water-cooled chiller system, which eradicates the need for human interventions under varying conditions. This was achieved by exploring the possibility of using the emerging field of Reinforcement-Learning controls, on one such chiller system. By taking over control of the cooling towers and the condenser pump, the algorithm aims at finding the optimal speed combinations for the immediate and near-future conditions, to reduce the overall consumption. The following sections present the development process of one such algorithm, starting by delving into the problem and looking at existing controls. Having chosen the desired algorithm, the development and implementation processes are presented, both from a theoretical and a practical aspect. The algorithm is then tested on a simulation model of the plant and also partially validated on the real equipment. The results obtained are discussed and compared to the existing controls, including analysis on whether the use of such algorithms is feasible in the real world.