PhD Oral Exam - Mirata Hosseini, Building Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Building energy simulation is commonly used to evaluate the energy performance of buildings to support decisions made at the design stage or to quantify potential energy savings of various strategies for retrofitting existing buildings. However, in many cases, the anticipated performance through simulation output significantly deviates from actual measured data. A major reason for such discrepancy is due to uncertainty in the simulation inputs.

One source of input uncertainty is weather data representing the climate condition. In order to predict the long-term performance of the buildings with energy simulation, modellers commonly use a single Typical Meteorological Year (TMY) weather data file which supposedly represents the climatic conditions. The single weather year file is composed of hourly resolution data from the most 12 representative calendar months of 30 years which are selected based on statistical similarity to long-term weather daily-averaged data. These weather files are synthetically constructed on historical weather data over a long period of time for an array of weather parameters, such as solar radiation, temperature, wind speed and others. The statistical procedure to construct the weather files depends on the weights assigned to these weather parameters. Under current practice, these weighting factors are universally assigned regardless of climatic locations nor the building application. This approach leads to energy performance predictions that deviate from the long-term averages.

Nevertheless, the single weather file ignores the variation in building energy performance resulted from natural weather variation. This source of uncertainty becomes even more critical when the long-term superimposed effect driven by human and anthropogenic factors are added to natural variation. Historical weather data shows that compared to other regions, higher latitudes, including Canada, have been affected more by climate change, and it is expected that this change will be even more in the years to come. Uncertainty due to weather variation and climate change is one of the main reasons for unexpected actual energy performance. Under the changing climate, building's energy performance is expected to change significantly in the northern climates, including Canada.
The current thesis mainly aims to address the two aforementioned issues with novel approaches:

1. Machine learning were deployed to extract the feature importance of the weather parameters in order to assign non-universal weighting factors straightly proportional to their impacts on energy performance of buildings. Weather files constructed with these systematically assigned weighting factors are climatic location and building type dependent. The newly constructed typical meteorological year weather files were applied to two different climatic locations to investigate the representativeness of these new weather files as compared to existing weather files and historical weather data of actual years. The representativeness was indicated in terms of the deviation in predicted energy performance of buildings between using the typical meteorological year weather file and actual historical weather data. The results indicated that typical meteorological year weather file based on the novel approach offers better prediction (with statistical significance) on energy performance for climatic locations with wider temperature range. As a result, the suggested method avoids potential under/oversizing of equipment and promotes energy conservation.
2. General circulation model (GCM) data considering various climate change scenarios based on socio-economic, population, land use, technology, and policies are used to provide information about future climatic condition. However, there are two primary challenges in application of data for building simulation:
   1. Bias in the models: considerable deviation can be found when the historical GCM data is compared to station observed weather data.
   2. Inadequate resolution: GCM data has daily temporal resolution rather than the hourly resolution required in building energy simulation.
      

In order to use this data for simulation purposes and better predict future building performance, further processing is conducted. A statistical bias-correction technique, known as the quantile-quantile method, is applied to remove the bias in the data in order to adapt GCMs to a specific location. The study then uses a hybrid classification-regression (K-Nearest Neighbour – Random Forest) machine learning algorithm to downscale the bias-corrected GCM data to generate future weather data at an hourly resolution for building energy simulation. In this case, the hybrid model is structured as a combined model, where a classification model serves as the main model together with an auxiliary regression model for cases when data is beyond the range of observed values. The proposed workflow uses observed weather data to determine similar weather patterns from historical data and uses it to generate future weather data, contrary to previous studies, which use artificially generated data. However, in cases where the future GCM data showed temperatures ranging outside of the observed data, the study applied a trained regression model to generate hourly weather data. The current study suggests a workflow that can be applied to global and regional models data to generate future weather files year by year for building simulation under various scenarios and, consequently, extreme weather characteristics are preserved for extreme or reliability analysis and design optimization.

In addition, a novel method is introduced to find building design solutions under uncertainty of weather variation and climate change. The design options are architectural and envelop features at different levels. A full factorial design of experiment is used for large-scale simulations and training deep neural network surrogate models to assess energy performance of design alternatives under multiple future years under various climate change scenarios. The method with application of a novel performance indicator is applied to explore design space and find the design solutions that most probably contribute to meet building energy performance targets over the project's lifespan.

This workflow takes into account the effect of weather variation under various climate change scenarios and suggests several design solutions that can be offered to stakeholders, architects, engineers, and third-parties including insurance companies. This way, design alternatives can be compared, and designs with a higher probability of success can be selected as a final solution. In addition, policy-makers can use the results and the suggested workflow to adopt and update national and provincial building energy codes such as National Energy Codes of Canada for Buildings (NECB) in line with the national policies following the Paris climate change agreement.