Non-Linear Autoregressive with exogenous input (Narx) chiller plant prediction model

• Main Authors: Azlee, Zabidi, Mohd Izham, Mohd Jaya, Wan Isni Sofiah, Wan Din, Hasliza, Abu Hassan, Ihsan, Mohd Yassin
• Format: Conference or Workshop Item
• Language: English; English
• Published: IEEE 2021
• Subjects: Q Science (General); QA76 Computer software; T Technology (General)
• Online Access: http://umpir.ump.edu.my/id/eprint/33474/1/Non-Linear%20Autoregressive%20with%20exogenous%20input%20%28Narx%29_FULL.pdf; http://umpir.ump.edu.my/id/eprint/33474/2/Non-Linear%20Autoregressive%20with%20exogenous%20input%20%28Narx%29.pdf; http://umpir.ump.edu.my/id/eprint/33474/; https://10.1109/ICSECS52883.2021.00077
• Institution: Universiti Malaysia Pahang

A chiller plant is a centralized system used for air cooling systems, commonly, for covering a large area of building with various components such as chillers, cooling towers, pumps, and chilled water storage tanks. Each component has several sensors or indicators with status information. Users can use the information to plan for maintenance and as guidance during troubleshot if an event occurs. It is crucial to ensure the chiller plant is operating efficiently without any faulty especially in critical buildings such as a hospital. The main problem of the chiller plant is to conduct preventive maintenance for avoiding the chiller plant failure and breakdown unexpectedly. Based on the literature, approximately 80 components in the chiller plant has found as the possible reason for the chiller plant faulty. In the current research, modeling chiller plants has been done by several researchers, objectively for preventative maintenance purposes. Study case for this project is for a chiller plant at Hospital Raja Permaisuri Bainun, Ipoh, Perak, Malaysia. A model for the proposed chiller plant system is to be designed using System Identification (SI) technique based on Nonlinear Autoregressive with Exogenous Inputs (NARX). Validation result shows, the proposed chiller plant system can be modelled and to be used as One Step Ahead prediction tool with residual Mean Square Error (MSE) of 1.018E-3 for training set and 1.017E-3 for testing set.