Multi-domain multi-objective optimisation of urban district environmental performance

Energy and environmental building performance simulation techniques have advanced considerably throughout the past half a century. Sophisticated and easy to use simulation engines now exist that can simultaneously simulate heat flow, fluid flow, plant systems, daylighting and radiation exchange and the impacts of building occupants on these systems and the corresponding feedbacks. Meanwhile, mass rural-urban migration has meant that the global population is now predominantly urban – we have become homo urbanus – with the vast majority, more than three quarters, of global resource use being concentrated into urban settlements. With growing concerns over climate change, there is thus a need to identify ways of reducing the negative environmental impacts of urban settlements, whilst ensuring that the quality of urban life is maintained or enhanced. This principle is enshrined in the UN Sustainable Development Goal 11: Sustainable Cities and Communities. In consequence, the last quarter of a century has seen a considerable increase in research activity to develop computational techniques that can efficiently and accurately simulate energy and environmental performance at higher spatial scales, and accounting for increasingly sophisticated building-occupant, building-building and building-system interactions. It is in this urban complexity that this thesis is situated.
Specifically, this thesis seeks to develop and apply a computational framework with which energy (using sunlight availability as a corollary), thermal and acoustic comfort can be simulated and optimised in the urban context, through a series of urban district use cases located in China. In the first instance, candidate building and urban morphology parameters are identified through cluster analysis. These parameters are then used to support parametric modelling of the above sunlight, thermal comfort and acoustic comfort performance domains, using dedicated simulation techniques. Hierarchical cluster analysis is also applied to these results to synthesise preliminary design guidelines. These simulation results are also used to train meta-models, developed using an adapted combination of generic regression neural network (GRNN) and grey wolf optimiser (GWO) algorithms. The trained and validated meta-models are then used to define the objective functions employed by a non-dominated sorting genetic algorithm with elitist strategy (NSGAII) optimiser to identify feasible Pareto solution sets of our three performance domains, employing the earlier defined morphology parameters. Finally, a specification for an interactive web-based urban design tool is outlined, with which the trained optimiser could be employed to suggest well performing morphologies as inspiration to urban designers.