Adaptive regression model-based real-time optimal control of central air-conditioning systems. (15th October 2020)

• Record Type: Journal Article
• Authors: Hussain, Syed Asad; Huang, Gongsheng; Yuen, Richard Kwok Kit; Wang, Wei
• Is Part Of: Applied energy. Volume 276(2020)
• Publication Date: 2020-10-15
• Subjects: Optimal control -- Data segmentation -- Adaptive model -- Energy saving -- Central air-conditioning system; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2020.115427
• Languages: English
• ISSNs: 0306-2619

Abstract

Highlights: Adaptive regression model-based real-time optimal control was developed for air conditioning systems. Regression models of the air conditioning process were adopted in this study. Similarity analysis was carried out to find the optimal length of training data. The proposed strategy is able to improve the building energy efficiency. The proposed strategy is able to reduce the computational load. Abstract: Model-based, real-time optimal control is an effective tool to improve the energy efficiency of central air-conditioning systems. However, its performance relies heavily on the accuracy of the system models, whereas the development of accurate models for central air-conditioning systems is not easy due to their complex dynamics and non-linearities. This study presents an adaptive regression model-based real-time optimal control strategy for central air-conditioning systems. In the proposed strategy, regression models are adopted to describe the relationship between the power consumption of the system and the variables that are optimised. Their simple structures enable a low computation load for model updating and real-time optimisation. The length of the training data (for model updating) is investigated, and a suitable length is found using a similarity check-based method. Case studies were carried out to assess the performance of the proposed strategy, and they demonstrated that (1) a week was the optimal length of the training data for the case system, (2) the proposed strategy saved energy use by 3.48–10.59% when compared with a benchmark system with no optimisation, and (3) the proposed method reduced the computational load by 85% when compared with a simplified physical model-based optimal control without adaptive modelling.