Use this URL to cite or link to this record in EThOS:
Title: Thermal behaviour model identification for three different office buildings
Author: Mustafaraj, Giorgio
ISNI:       0000 0001 3437 6012
Awarding Body: Brunel University
Current Institution: Brunel University
Date of Award: 2008
Availability of Full Text:
Access through EThOS:
Access through Institution:
The thermal behaviour was investigated of three offices positioned in three buildings built in different periods, one academic institute built in 1920 and two modem commercial buildings in London. The buildings chosen for this study are the Rockefeller Building, which is part of University College London (UCL), Portman House in Oxford Street and the Visa Building in Paddington. Due to the lack of specific information related to the structure of the buildings such as windows, doors, building dimensions and other information that would allow the use of physical models, in this project black-box linear and non-linear mathematical models were used. Data relating to room temperature, hot and chilled water temperature, air flow and temperature from air handling units and outside temperature were collected for one year, from the actual building management systems (BMSs) installed in these buildings. The main assumption of the model development in the three buildings was that although occupancy, computers, printers etc cause an additional internal heat gain, their impact is in part indirectly included in the model. The primary objective of the analysis was to identify the inputs (independent variables) that gave good models for the prediction of room temperature for a certain period. Consequently, the process of input selection and period of validity in obtaining models that give good thermal prediction (within the same period) were the key points in season subdivision. The first part of the analysis applied the following linear parametric mathematical models to the three office buildings selected: Box Jenkins (BJ), autoregressivem oving averagew ith exogenousi nput (ARMAX) and output error (OE) structure. The project then deals with non-linear mathematical models. The same inputs selected and assumptions made with linear analysis were used to build, in turn, models with feedforward backpropagation (FFBP), non-linear autoregressive mathematical models with parallel arrangement (NARX) and series-parallel arrangement (NARXSP). The research presented in this project is related to developing models for three real offices positioned in three different buildings whereas previous researchers have applied these models mainly to experimental rooms and HVAC plants, with the purpose of fault detection and diagnostics. Furthermore, in the past, research on thermal model development has been related to one office or HVAC plant, and for a limited period of time (a few weeks or months). In contrast, this study undertakes an overall analysis of thermal model development for three offices and for a period of one year, where the process of input selection is given priority to obtain good models. Thus, previous studies have not utilized these two types of models for such a long period of data collection nor related them to three different buildings. Finally, model development and then validation were pursued utilizing the same week, different weeks and different days (where the first part of the data in each case was used for model estimation and the following part for model validation). This was one within the period that the models gave good results for the prediction of room temperature. The best mathematical models (linear and non-linear) that predict the room temperature, in terms of the inputs selected, has been determined for each season. The procedures for how to choose the best models are based on the following techniques: final prediction error (FPE for linear models), mean squared error (mse for non-linear models), and model fits and errors between measurements and simulated model output. Overall, the results related for the prediction of room temperature with non-linear models, are better than those obtained with linear models, as a result of comparison between models' errors, FPE and mse obtained with linear and non-linear models.
Supervisor: Chen, J. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available