Intelligent control system design for energy conservation in commercial buildings

Abstract

This thesis focuses on the development of model predictive control (MPC) strategies for reducing energy consumption in air-conditioned buildings. It is well known that the building sector is responsible for 40 per cent of the world’s energy usage and 33 per cent of all greenhouse emissions. As a result of global environmental issues and decreasing energy resources, there is strong motivation to develop more efficient control strategies for Heating, Ventilation, and Air Conditioning (HVAC) systems in buildings. The existing HVAC control strategies are not energy or cost efficient, which results in energy waste, high on-peak electricity demand and poor thermal comfort in buildings. Previous works have shown that MPC can be utilised as a supervisory controller to achieve energy saving while maintaining the indoor thermal comfort in buildings. However, most of the past studies were focused on small residential buildings or mid-size commercial buildings. It is highly desired to improve the existing MPC strategies to make them more reliable and applicable for large commercial buildings. This thesis extends the previous works by addressing the following challenges when dealing with the large buildings. Firstly, HVAC plants and the thermal dynamics of buildings are inherently nonlinear. Accurate modelling of these components is difficult due to the limited number of sensors that are usually installed and the paucity of prior knowledge of the system. There is a need to develop models that are capable of effectively handling the nonlinearity to achieve better modelling accuracy. Secondly, in large commercial buildings with adjacent large open spaces, the effects of thermal coupling between differently controlled spaces play a crucial role. This significance of the interaction between zones has seldom been discussed before and requires more thorough investigation. Thirdly, although load shifting function of MPC have been proven to be effective in achieving cost savings in buildings with a considerable thermal mass, it is demanding to investigate the application value of these strategies in lightweight commercial buildings. Finally, given the presence of uncertainties, these models may not be able to predict the indoor temperature accurately, which may lead to poor control performance and even instability in operation of the MPC strategy. The existing robust control approaches are generally too conservative, and may not be suitable for use in real-world buildings. In this study, the advantages of neural networks (NNs) will be exploited to address the challenges outlined above. NNs are known as universal approximators, meaning that they can model any continuous functions with any desired degree of accuracy. In particular, the NNs will be used to conduct modelling work, generate control rule, and improve the performance of classical MPC. The major contributions of this thesis are presented in four chapters, with each based on an individual scientific paper. Paper-1 presents a systematic modelling method for air handling units (AHUs) and thermal zone using a recursive NN (RNN). As the major novelty, a cascade NN structure is developed, which enables the thermal dynamics modelling of both interior zones and perimeter zones within investigated building. This approach allows accurate prediction of both supply air temperature and zone temperature prediction, making it suitable for predictive control design. Continuing with the first paper, Paper-2 introduces a multi-input, multi-output (MIMO) model, which effectively models the convective heat exchange between open spaces within multi-zone commercial buildings. The proposed model allows closed-loop prediction for several adjacent zones simultaneously by considering their thermal interaction. A NNbased optimal start-stop control method is also developed in this paper to demonstrate the energy saving potential enabled by using the proposed predictive model. The NN models provide accurate prediction results, but they are in general difficult to optimise under an MPC framework. Paper-3 presents a hybrid MPC (HMPC), which combines the classic MPC with an inverse NN model. With the HMPC, the classical MPC based on linearised building model optimises the supplied cooling energy. The inverse NN model compensates the nonlinearity associated with the AHU process and generates more accurate control inputs. Simulations and experiments demonstrate the feasibility of the proposed method in achieving energy and cost reductions while maintaining good indoor thermal comfort in the investigated large commercial building. The MPC formulation in Paper-3 does not take the system uncertainty into account. In reality, however, the modelled building energy systems are always affected by uncertainties, so that the modelling errors become inevitable which cause control performance degradation to the MPC. Paper-4 considers the application of a robust MPC (RMPC) to handle the system uncertainty within buildings. In particular, an uncertainty estimator is developed based on the previously presented RNN model to provide uncertainty bound to the conventional closed-loop min-max RMPC. The newly developed bound estimator reduces the conservatism of the RMPC and achieves improved control performance. In conclusion, the research work presented in this thesis has made important contributions to the research of intelligent model predictive control for air-conditioning systems in commercial buildings. The methodologies developed in this thesis can be utilised for other buildings or for the control of other dynamic systems.