UBC Theses and Dissertations
An integrated fire risk management of multi-unit residential buildings with smart and green features : machine learning-based framework Ouache, Rachid
The building industry has been facing significant challenges, including high quality, sustainability, cost-effectiveness and safety. Fire risk poses a significant threat and impacts on public safety, property, and the environment. Canada recorded 0.5 million fires in 10 years, which caused 15,000 fatalities, a direct loss of $7.5 billion, 1,200 children were injured, and 1.2 million Canadians were affected directly. In British Columbia (BC), 55% of the fires were associated with multi-unit residential buildings, which are being increasingly built with smart and green features for a positive impact, however, these features may negatively impact from a fire’s perspective. This research developed an integrated fire risk management framework using machine learning algorithms for multi-unit residential buildings integrating smart and green features (SG-MURBs). Five phases were incorporated in the framework to enhance fire prevention, protection, and intervention strategies: (I) investigated state-of-the-practice to explore fire risk management practices (II) Identified key potential fire risk and safety-related factors considering relative frequencies and their related impacts (III) Developed benchmarks to set the acceptability levels and determine the critical factors accordingly (IV) Developed models to predict the potential impacts of fire incidents in SG-MURBs, and finally (V) Generated optimal solutions and optimized multidimensional fire impacts. The developed framework was applied and tested for seven cities in BC, Canada. The results discerned the key potential contributing factors to fire incidents covering common, smart, and green factors, including 40 ignition sources, 28 human errors, and 36 combustible materials. Using the developed benchmarks, Vancouver, Kelowna, and Kamloops were found the top critical cities with 58%, 37%, and 47% of very high levels of ignition sources, human errors, and combustible materials, respectively. The most suitable methods for fire control and fire suppression materials for the top critical combustible materials were determined. The ANN models were found to outperform classifiers with prediction abilities of 65%, 72%, 85%, and 99% in predicting related potential impacts. Moreover, the optimal set of fire safety strategies were determined using genetic algorithms to enhance fire risk management. The developed framework will help decision-makers guide policies and enhance investments for specific fire prevention, protection, and intervention strategies accordingly.
Item Citations and Data
Attribution-NonCommercial-NoDerivatives 4.0 International