Urban fabrics as a tool of urban performance mitigation: Algorithmic approach for bridging building geometry with urban performance optimisation

Lila, Anas

Abstract

The advancements in methods of built environment design have led to the rise of computational methods in urban modelling and environmental simulation to aid the early stages of the design process. Computational urban modelling and simulation methodologies can use a parametric approach to enable geometrical dynamic modelling and investigate urban environmental performance. This thesis aims to understand the effect of urban geometry on urban performance in an algorithmic approach to reach for an efficient optimisation approach based on this impact. To achieve this goal, a framework was developed to enable a time-efficient performance-based optimized design to guide the urban design process in the early stages. This framework establishes a new environmental data-driven approach for designing urban neighbourhoods. A preliminary sensitivity analysis measured the relative importance of geometrical variables, their impact on performance aspects and the computational time. It was conducted in two locations, Aswan, Egypt, London, UK, for cooling and heating demands. The results of this analysis quantified relative importance for the tested geometrical variables’ impact on energy demand. It was clear that computational and time cost is limiting the capability of conducting general performance optimisation on the urban scale. This led the research to the classification of geometry to optimise urban geometry based on its solar radiation performance. The parametric workflow presents a methodology to break down the neighbourhood model into its geometrical variables: location, orientation, building’s area, height, typology, and the surrounding geometry context. Then, a database of text annotations for generated buildings was attached to its solar radiation simulation results. These annotations are used as indicators to match the following geometry generations to save simulation time in similar geometrical scenarios. The framework was used to optimize solar radiation for a neighbourhood geometry in Aswan, Egypt. Machine learning principles were adopted to provide the framework with prediction capabilities of solar radiation performance with accepted prediction accuracy and reduced time consumption. A positive linear correlation was found between machine learning principles and its equivalent simulation results for architectural and urban scales. The proposed prediction approach succeeded to achieve significant time savings compared to the traditional simulation process with acceptable accuracy. These reported findings shed new light on the capability of optimisation in the early design stages. The Genetic Algorithm’s optimisation principles show a significant capability to find optimal or near-optimal solutions for hypothetical and existing neighbourhood context tests while saving more than80% of the computational time needed. These results present a template for using data-driven urban design to inform environmental decisions in the early design stage at the neighbourhood scale

• Thesis Type: Thesis
• Deposit Date: Apr 26, 2022
• Public URL: https://uwe-repository.worktribe.com/output/9192425
• Award Date: Nov 6, 2021