Data-driven Modelling and Real-time Optimization for Diversified Chiller Plants

Project Key Words:: Machine learning, chiller plant optimization, data driven energy management system

Project Description:: Modern chiller plants are controlled by fixed rules using logic controllers (PLCs or DDCs), sensors, actuators, electrical systems, ACMV systems etc. Modern rule-based Energy Management Systems (EMS) developed by domain experts from the world’s leading BMS suppliers are known to provide unsatisfactory results during the operation and maintenance phases.
The objective of the project is to develop a highly dependable data driven EMS that learns from sensor data to improve chiller plant design, diagnostics, optimization and control.
The existing physical models over chiller plants are not very useful for energy optimization, because of 1) the high dynamicity of the system; 2) long response latency of control actions; 3) real-time constraint on the optimization; 4) poor scalability of successful experience to new plants.
This project adopts both the state-of-the-art deep learning and simple machine learning techniques to enable accurate modeling of large number of chiller plants, regardless of brand, model and age of the equipment used in the chiller plants.
The machine learning models are implemented as energy management systems that are highly resilient, dependable and self-learning for 15-25 year operating life cycle.

Project Impact: Deployed at overseas (India) and local pioneering projects at Racks Central Data Centre and Keppel Bay Tower