Use this URL to cite or link to this record in EThOS:
Title: Control of naturally ventilated buildings : a model predictive control approach
Author: Sykes, Joshua S.
ISNI:       0000 0004 6060 7306
Awarding Body: University of Sheffield
Current Institution: University of Sheffield
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
During operation, buildings consume a large amount of energy, around 40\% of global final energy use. A major challenge is to reduce the amount of energy used while still providing a comfortable environment for building occupants. The use of passive techniques, such as natural ventilation, is promoted in certain climates to provide low energy cooling and ventilation. However, controlling natural ventilation in an effective manner to maintain occupant comfort can be a difficult task, particularly during warm periods. One area which has been identified as having the potential for reducing energy consumption while maintaining occupant comfort is the use of more advanced control techniques. A technique which has been much explored in recent years for application in mechanically ventilated buildings is Model Predictive Control (MPC). MPC is a control technique which uses a model of the system dynamics and by solving an optimisation problem is able to determine the optimal control inputs. In this thesis the application of MPC to naturally-ventilated buildings is investigated. The essential component of an MPC strategy is the predictive model of the building's thermal dynamics. An empirical approach to modelling was taken using multilayer perceptron (MLP) neural network models. To use empirical data from a building to create a predictive model it is essential to ensure the quality of the data is appropriate. In order to assess the data available from buildings during normal operation four studies were carried out in different buildings. The data collected from these studies represent a range of natural ventilation scenarios and building types in different locations in the UK. To test the impact of identification procedures upon the resulting neural network models, an identification experiment was carried out using dynamic thermal simulation. Neural network models were trained using both the data from real buildings and the simulation data. Results showed that neural network models trained using data from real buildings were capable of good predictions. However, the lack of input excitation during normal operation resulted in models which did not capture the effect of the window opening control. The identification experiment demonstrated that by exciting the control input the resulting neural network models captured the effect of the control, making them suitable for MPC. The main focus of this thesis is the investigation of techniques to develop predictive models which can be utilised as part of an MPC strategy. However, to demonstrate the potential benefits of MPC a controller designed to maintain a suitable internal temperature is demonstrated. The controller utilised the neural network models developed using the data from the system identification experiment and a non-linear optimiser. The MPC method showed the potential to reduce overheating and improve upon the typical control used in the majority of buildings. Findings in this thesis demonstrate that empirical models capable of good predictions can be trained and could be successfully applied to the control of natural ventilation systems. Furthermore, the potential advantages of adopting an MPC approach to natural ventilation control are shown.
Supervisor: Hathway, Abigail ; Rockett, Peter Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available