Solar potential in early neighborhood design : a decision-support workflow based on predictive models

In light of the acknowledged need for a transition toward sustainable cities, neighborhoods and buildings, urban planners, architects and engineers have to comply with evermore demanding energy regulations. These decision-makers must be supported early-on in their process by adequate methods and tools. Indeed, early-design decisions, which concern parameters linked to the building form and urban layout, strongly dictate the solar exposure levels of buildings, in turn influencing their energy need (e.g. for heating and cooling) and production potential (e.g. through on-site active solar systems). Despite the spread of existing digital tools, limitations remain, withholding their integration into the early design process. These considerations lay down the context within which this doctoral research was carried out. The main objective of this thesis is the development of a performance-based workflow to support decision-making in early-design neighborhood projects. The performance is here defined through three criteria: (i) the daylight potential, quantified by the spatial daylight autonomy, (ii) the passive solar potential, quantified by the annual energy need for space heating and cooling, and (iii) the active solar potential, quantified by the annual energy production. The research process consisted of two main phases. First, the development of a performance assessment engine allowing real-time evaluation of an ensemble of buildings. Second, the integration of this method into a decision-support workflow, taking the form of a digital prototype that was tested among practitioners. For the first phase, a metamodeling approach was adopted to circumvent the limitations associated to simulations involving solving physics-based equations. Mathematical functions were obtained to predict the daylight and energy performance of a neighborhood, from a series of geometry- and irradiation-based parameters, easily computable at the early-design phase. To derive these functions (or metamodels), a neighborhood modeling and simulation procedure was executed to acquire a dataset of reference cases, from which the metamodels were trained and tested. The resulting multiple-linear regression functions, combined to an algorithm for quantifying the active solar potential from the irradiation data, formed our performance assessment engine. To assess its usability and relevance, the workflow was implemented as a prototype, supported by existing 3D modeling and scripting tools. Inspired by the emerging performance-driven and non-linear design paradigms, a multi-variant approach was adopted for this implementation; from the space of possible designs defined by a small set of user-inputs, a series of neighborhood variants are generated through a random sampling algorithm. Results of their evaluation by the core engine are displayed to allow a comparative assessment of the variants in terms of their morphology and solar potential. Having been tested among practitioners during