Chiller energy consumption prediction : methods and system development

• Main Author: Xiao, Qianru
• Other Authors: Ho Shen-Shyang
• Format: Final Year Project
• Language: English
• Published: 2016
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Simulation and modeling; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: http://hdl.handle.net/10356/66778
• Institution: Nanyang Technological University

Power consumption is critical in organization operations, for various reasons including financial saving and environment protection. Power usage prediction helps organization to understand how energy consumption is affected by different factors, and therefore provides decision-making support for further energy optimization. Many organizations use historical consumption pattern to estimate future trend, which lacks accuracy and credibility. Internet of Things and data analytics offer solutions to address this problem by collecting and modelling large amount of data. This project deals with the field of hourly chiller power consumption prediction by building up an end-to-end analytical system for an energy consulting start-up company. From raw data collected by local weather station, and readings fetched by sensors installed at different parts of chiller plant, meaningful information are extracted and pre-processed. After exploring the features of data, different predictive techniques including linear regression, support vector regression and artificial neural network are examined on sample data to determine the best learning method. In addition, this project also investigates if introducing time-related variables has significant improvement for developing the model. On top of the model, accumulative training and sliding window training techniques are also compared to give the model the ability to adaptively learn about new data. To provide company the transparency, the ease of using analytics and the automation of processes, the front layer is implemented as a web-based java application where user could validate model and test new data. The system is developed with various open source technologies including R, Java, Bootstrap, Data-Driven Documents, which not only brings the company a cost cutting solution but also maintainability and sustainability. With Data Centered Architecture and Model View Controller Design, each component of the system is loosely coupled with each other so it can be easily changed and reused in future development.