Model predictive control for energy efficiency and occupant well-being optimisation in tropical buildings

Conventional building automation and control (BAC) systems employ reactive feedback control, such as proportional-integral-derivative (PID), which limits the efficiency in managing building energy and dealing with the contrasting need for human well-being and energy efficiency. Model predictive control (MPC), which brings forward prediction and optimization capabilities on the principles of closed-loop control to BAC, gains extensive attention as a solution to such limitations, allowing optimisation of building energy efficiency and indoor climate. This study proposes an MPC system based on a building-physics-based modelling approach to formulating a state-space model (SSM) that captures the building dynamics and quantitative occupant comfort indices. The system incorporates real-time optimization for energy efficiency and occupant comfort, which requires less than 0.1 s for solving the optimisation problem in each control interval. The proposed MPC system was implemented in two test buildings, the Building and Construction Authority (BCA) SkyLab and Lecture Theatre 3 (LT3) on the campus of NTU. The control characteristics and performance of the MPC system were evaluated and compared to conventional BAC systemS under different air-conditioning and mechanical ventilation (ACMV) systems including fan coil unit (FCU), active chilled beam (ACB), air-handling unit (AHU) as well as separate sensible and latent cooling (SSLC) with dedicated outdoor air system (DOAS). The MPC system achieved 15 – 20% of electricity savings for the ACMV systems with improved occupant comfort, compared to conventional BAC systems. The MPC system was further developed to perform coordinated control of multiple building systems (air-conditioning, lighting and shading) by incorporating extra models for predicting indoor lighting power and visual comfort. The MPC system achieved up to 20.3% of building electricity savings with great improvements in thermal and visual comfort. In order to enhance the adaptability of the MPC system to the transient nature of building operation, machine learning (ML) technology was adopted to develop an adaptive ML-based building model, which was initialised by historical building operation data and continuously updated using online building operation data. An adaptive MPC system employing such model was developed and implemented in a test office. The MPC system achieved 59% cooling energy saving compared to the conventional thermostat control. This study also explored the feasibility of developing a full ML controller by supervised learning of the control laws of the MPC system using the building operation data generated by the MPC system. The ML controller retained the high thermal comfort performance and 86% of the energy saving performance of the master MPC system.Doctor of Philosoph