Aprender, simplificar, acelerar: machine learning para a acessibilização de análises em escala urbana

• Ano de defesa: 2021
• Autor(a) principal: Passos Filho, José Aderson Araújo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: UFC
• Palavras-chave em Português: Planejamento urbano; Conforto térmico; Machine learning
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/58871

This work proposes the implementation of Machine Learning to simplify and make accessible the obtaining of complex analyses’ results, more specifically in the assessment of thermal comfort on the urban scale. The complex relationship established between planning, the shape of the city and climate conditions makes it necessary to use strategies to analyze and produce urban space that often exceed the planner's theoretical and technical expertise. In order to improve health and increasethe social life of its inhabitants, the local climatic characteristics of a city must be considered in order to promote its environmental comfort. This is added to the inherent complexity of large urban centers, which are increasingly occupied in comparison to their rural counterparts, and of the tools available for analyzing their phenomena. It is therefore important to think about the trade-off between precision and speed ofmethods applied to the construction of tools that are not only potent, but that can also facilitate a quick and constant action of planning professionals — initial design stages accept a type of feedback that is faster and less accurate, while there is still more room for exploration. Since architecture has traditionally been a discipline almost entirely devoid of rigorous data analysis, it is, however, possible to realize that “data” is increasingly becoming a protagonist element in interactive design. In particular, and pertinent to this work, urban planning, supported for a longer time by data analysis, manages to become even more perfected by this same trend. It is, then, that, for the estimation of results of thermal comfort analyses on the city scale, the proposed method substitutes calculations of solar geometry and computational fluid dynamics (CFD) simulations with models simplified through Artificial Neural Networks (ANN), trained with density measurement data from different levels of urban aggregation obtained through a geographic information system (GIS) and parametric modeler on a computer aided design (CAD) platform, in correlation with a percentage indicator of thermal comfort calculated according to hundreds of simulations of shading and natural ventilation. Thus, interfering in thermal comfort is not just up to the architect of intralot spaces, especially when talking about the consequences of urban form for extralot spaces. There is a need and, with this method, the possibility for the planner to act consciously on this aspect of urban performance, in a simple and accessible way. From a technological, political and environmental perspective, the proposal aims, finally, through the construction of information, to improve the understanding of the implications that buildings bring to the urban environment and to contribute to the production of the contemporary city.