Overview
Chiller plants in buildings consume the greatest electrical power in Air-conditioning system, playing a pivotal role in our journey towards carbon neutrality. Chiller plants have conventionally been controlled using traditional rule-based strategies, resulting in energy inefficiency and limiting system adaptability to environmental changes. This paper reveals the success of implementing Artificial Intelligence-based model for large-scale real-time monitoring and control of chiller plant which seizes every opportunity to enhance building energy performance.
Context
With the introduction of modern high-efficiency chillers and a central Regional Digital Control Centre, there is an opportunity to implement control strategies to achieve energy saving by varying Coefficient of Performance under different part-load and ambient weather conditions. To acquire optimized chiller parameters, there were past trials using Genetic Algorithm (GA), Particle Swarm Optimisation (PSO), or combination of both in achieving optimisation of engineering systems.
Purpose
This study aims to develop a scalable system, code-named ‘ChillStream’, based on the novelty in Artificial Intelligence (AI) chiller optimisation. Artificial Neural Networks were trained using historical plant data and weather data to predict power consumption and cooling load for individual chillers. In an attempt to combine the merits of evolutionary algorithm and swarm intelligence, a hybrid GA-PSO Algorithm was developed to calculate optimised setpoints at regular time intervals.
Approach
The developed AI control strategy was successfully deployed in a chiller plant with a significant cooling capacity installed in a clinical laboratory building in Hong Kong. Compared to the conventional rule-based system control, the chiller plant’s overall energy consumption was prominently reduced by 8% in autumn/winter 2023.
Insights
Through the autonomy of ChillStream to operate the chiller plant, considerable manpower resources are saved. This optimisation control strategy can be readily replicated and adjusted to accommodate the unique configurations of chiller plants in various buildings, resulting in substantial energy savings.