Research On Surrogate-assisted Optimization Methods For Facilitating Ventilation In Urban Blocks

In China’s rapid urbanization over the past 70 years,how to balance urban development and environment protection has been one of the major concerns for planners in China.For its close associations with air pollution,residents’ well-being,building energy consumption,and human thermal comfort,urban ventilation plays a pivotal role in determining the quality of urban environment.That said,it is necessary to propose some universal and evidence-based design optimization methods,to facilitate flexible,fine,and standardized wind environment quality management for cities in China.This thesis conducts multi-disciplinary research by synthesizing knowledge from various fields including urban planning,meteorology,operations research and management,statistics,etc.,to develop several surrogate-assisted design optimization approaches for improving urban ventilation capacity at block scale in Guangzhou.The thesis begins with a brief introduction of the surrogate-assisted design optimization framework.In practice,designers often employ wind tunnel experiments and computational fluid simulation(CFD)models to study wind regime in urban blocks,with an aim to seek optimal plans that facilitate ventilation in complex urban areas.However,the costs of wind tunnel experiments and CFD simulations are usually prohibitively high,preventing designers from enumerating sufficient solutions before reaching a satisfying decision.The surrogate model is a model that approximates the relationship between urban morphological parameters and wind performance indicators.It is recognized as a cheap replacement for the CFD simulation with an ability to achieve fast examination and comparison of candidate solutions.When combined with a heuristic search algorithm,the surrogate-assisted optimization method can further help designers quickly detect near optimal design solutions at local or even global level.There are four functional modules incorporated into the proposed design optimization framework: a design of experiment module for sampling planning,a performance examination module for wind simulation,an approximation module for surrogate construction,and an optimization module for design optimization,with each module embedded by different tools and strategies.Generally,in a ventilation design optimization problem,there are design objectives,design variables,and design constraints.First,designers may define the design objectives according to several criteria,i.e.,the local climate characteristics,the interests of different stakeholders,the spatial context of the research area,etc.And they should carefully choose the design variables that can significantly influence the airflow regime in the urban area.Moreover,some morphological indicators specified in the planning regulations,such as floor area ratio and building coverage ratio,should be considered as design constraints which aim to place limits on certain combinations of design variables in the design optimization problem.Moreover,since it is hypothesized that the best surrogate approach will vary according to different ventilation optimization instances and applications,we propose a tree structure diagram to classify a design problem into different scenarios given its quantifiable aspects,e.g.,nonlinearity,the number of its design objectives and design variables.This information is then used to create "rules of thumb" dictating under which conditions each approach under different functional modules is most applicable to the design of an urban wind environment improvement project.This thesis utilizes three case studies to examine the efficiency and usability of the design framework under different research backgrounds.In the first case study,we present a surrogate-based optimization method to devise near optimal designs that minimize a building’s adverse impacts on nearby pedestrians’ wind comfort in an old neighborhood in Guangzhou.Four design parameters,i.e.,building width,building depth,building height,and building orientation,are considered for a greatly simplified building design task.Specifically,CFD analyses are performed to calculate the mean wind velocity,while all CFD experiment samples are determined through the BoxBehnken design of experiment method.Based on design samples,the relationships between building design variables and the summer and winter mean wind velocities within a specified assessment area are learned using the response surface methodology.Finally,a genetic algorithm with the desirability function as the objective function is applied to identify near optimal design options under the robustness control for response surface models.Next,based on the same studied case and by using the wind speed exceedance probability as the wind comfort assessment standard,we integrate Gaussian process(GP)regression models and the Latin hypercube sampling method to approximate the relationships between design variables and wind velocities at locations of interest.With the objectives of maximizing the winter and summer wind comfort and minimizing the modeling errors,robust and optimal designs of a target building in the infill development project are explored using the multi-objective evolutionary algorithm.And finally,we discuss the wind environment optimization task in a spatial planning problem of a newly developed residential block,where an advanced surrogateassisted evolutionary optimization algorithm is introduced to detect optimal design alternatives that maximize the project’s development profits and facilitate a comfortable wind environment.After generating 400 sampling plans using a parametric design module,we apply the CFD model to examine the samples’ wind behaviors and employ the Gaussian regression and gradient boosted regression tree regression to approximate the coupling relationships between morphological parameters and wind performance indexes.Optimal plans are obtained using the hierarchical surrogate-assisted evolutionary optimization algorithm.Through case studies,it is confirmed that the surrogate-assisted design optimization framework has universal applicability in design problems with different number of design variables and various design purposes.Moreover,by decomposing an urban ventilation design problem into different scenarios based on the classification diagram,we are able to select proper tools to formulate a working scheme proposed specifically for a unique wind environment optimization task.And meanwhile,it is believed the optimal plans derived from every wind environment optimization task can inform and inspire many other ecologically-oriented urban design practices.