An machine learning and urban morphology Based approach for AUTomatic generation of sustainable layout ## abstract 在城市规划中协调自上而下的政府规划和自下而上的设计实际问题是非常困难的 How to 自动化 GENERATE new urban design in the city context in a way that the new urban form improve the entire urban Synergy and its sustainability is the main question to be discussed in this paper. Study of urban form is an important area of research in urban planning/design that contributes to our understanding of how cities function and evolve. However, classical approaches are based on very limited observations and inconsistent methods. 快速的城市生成和预测的问题研究，一方面对于城市形冭学量化的研究算法局限于算法和生成的城市往往只考虑单独的元素而不是城市语境 ，另一个方面新的ai技术生成城市，人们难以有具体指标去控制最后的成果 This paper proposes deeplearning based approach to generate considering urban contexts, rather than individually analyzing the urban elements. We created data including the morphological characteristics of the buildings and their surrounding urban spaces. By 训练the data, statistical techniques identified them pathological features and types to 交互式的 generate city 以米兰为例 城市形态转型。 How to 快速的生成urban design in the city context in a way that the new elements improve the entire neighborhood 舒适 performance and its sustainability is the main question to be discussed in this paper 以米兰做运河地区做实验。自动的在osm上找到了2000个满足标准的样本。gaugan 模型和分布式训练，分别训练了两个机器学习数据库，一个用于路网和形态边界的生成，一个用于建筑元素的生成。设计师布局控制形态的关键因子地块开口和内部空地， 就可以快速的生成多种城市的结构和建筑3d精细模型排布来寻找城市设计的解决方案来解决 米兰大量的量化形态学研究和大量样本研究，快速的 以 ，功能混合度，地块密度，交通可达性，平均高度，多各方面来验证城市生成工具的适用性和城市结构预测能力. 发现同比于原来的地块 ，ai生成的方案很好的解决了城市密度和功能混合度问题，相比于传统算法更多的适应性和方案可能性，但是对于交通能力和原有地块相比类似。 This paper reports on the analytical potential of machine learning methods for urban design generation based on Therefore, this research aims to apply GAN in creating urban design plans, helping designers automatically generate the predicted details of buildings configuration with a given condition of cities. The proposed methodology was tested on the French Riviera and outputs show a moderate predictive capacity (i.e., adjusted R2 = 0.75) and insightful explanations on the nuanced relationships between selected features of the urban environment and street values. These A novel index termed Piping Coverage Rate was jointly proposed to evaluate the obtained results. The system produces the output within 45 seconds, which is drastically faster than the conventional manual workflow. And can we integrate physical and non-physical elements of spaces in one coherent informational space? abstract Study of synergetic urban planing（sup） is an important area of research in urban morphology form that contributes to A SUSTAINABLE URBAN transformation. 尽管SUP 很重要 classical approaches are based on very limited observations and inconsistent methods. 没有大量的量化urban 形态学form data aisist urban design ,生成考虑城市生成问题只有人工有限样本的城市形态学的定量单独分析 however ,only the feature metrics of individual urban artifacts and hence fail to capture contextual and relational aspects of the city. 诸多城市形态学关注的都是城市单独元素而不是城市的完整系统。 how to 应用整体性的方法 处理大量的urban morphology character and morphology structure assist Synergetic urban plan generation in a fast way 是一个难题 computeral自动化的城市分析尤其是基于deeplearning的方式能够大量样本的快速学习城市特征快速处理复杂的数据的讯息。这为整体性的方法生成sup设计提供了可能性。 However, their research focus on analysis smilar and 聚类大量的城市形态特征 还没有 根据具体的城市语境生成建筑排布，形成tool that asisst designers 处理大量的城市样本形态和快速交互的响应式辅助来生成一个城市设计 abstract 2 The difficulty with urban transformation is that designers need to analyse a large number of case studies of the relationship between complex urban forms and corresponding performance , and design strategies based on perceptions. This approach has a small sample size and is not accurate 1 The traditional method requires artificial perception and analysis of a large number of morphological forms, with a small sample size and inaccurate perception 2 It is difficult to use a holistic approach to urban planning to synergise the various stages of the project and the various roles This paper proposes an interactive way to automatically generate sup by deep learning model and morphological rules considering a wide range of urban contexts in a holistic way , rather than a single urban element Collection of data form osm data platform including the morphological characteristics of the buildings and their surrounding urban structure types. By training the data baed gausgan model that automatically learns the hierarchical structures of urban forms and fast provide the urban plan generation in a few second The following sections describe our method in technical detail and illustrate it through a case study in the city of milan. Synergy factor such as Functional mix degree, public space accessibility, building density, etc. urban index was jointly proposed to evaluate the obtained results compared with Current plan。 and reflect on some limitations and avenues for future work INTRRODUCTION promblem ？ The study of urban morphology plays an effective role in the process of sustainable urban transformation. However, traditional methods are still difficult to clearly assess its accuracy because the designer’s intuition dominates the design process. Designers have to gather and analyse a number of samples artificially, then spend much time working on sketching for different characteristics of urban design in the initial stage. The rapid emergence of quantitative analysis methods enhance its credibility in recent urban morphology research. However, they cannot solve the problem of lacking a large collection of professional samples, which leads to the designer's guesswork when measuring analysis indicators rather than summarizing reasonable thresholds from existing self-consistent designs. In addition, the introduction of machine learn-ing algorithms provides the possibility to analyze and generate design schemes more efficiently. Despite of this, existing methods are not combined with professional urban design analysis methods organically, resulting in the schemes are generated only from a statistical perspective but lack of professionalism. To sum up, current quantitative studies and machine learning methods suffer from the main limitations of lacking samples with high urban context fitness, lack of interpret-ability and all-in-one research tools. In this paper, an interactive method to automati-cally generate urban planning scheme by deep learning model combined with morpho-logical rules considering a wide range of urban contexts in a holistic way, rather than a single urban element. The contributions of this paper are: Synergetic urban planing (sup) background 形态学的关键要素影响城市转型 Study of synergetic urban planing（sup） is an important area of research in urban morphology form that contributes to A SUSTAINABLE URBAN transformation. 更大的范围很迷惑 According to the relevant theories of the British Konzan School, which is the most representative in traditional urban morphology research 2017 Moosavi, Vahid Within the domain of urban studies, urban morphology is the classical study of urban forms and their underlying formation processes and forces over time urban planning require detailed information about the functional, morphological and socio-economic structure of the built environment. The building stock directly affects urban structure, e.g. urban form, density of housing and distribution of population. (2013Robert Hecht*) A number of papers discuss which factors of urban morphology play an important role in urban regeneration and sustainable urban performance 核心 According to the relevant theories of the British Konzan School, which is the most representative in traditional urban morphology research: the key elements of urban morphology include the urban plan (streets, plots and buildings), the architectural composition and land use urban morphology character This city can comprise different kinds of urban artifacts (Rossi 1982) - human- made objects such as buildings, infrastructures, public space, etc Aldo Rossi [4] and Frey Hildebrand [5] believe that city’s form, and its morphology, are derived from their constituent form and physical assessment. The The theory of the sup city has been developed by Jane Jacobs on urban diversity [6], Jan Gale on the vitality of street space [7] and further discussions by many urban design researchers and practitioners [ Most of the principles of urban spatial vitality can be summarised in three key elements of urban form: good street accessibility, appropriate building intensity and building form, and sufficient functional mix (Table 1). 2016 Yu, Ye urban morphology character ![](https://www.filepicker.io/api/file/2P3tGVcvTEiza42N5Kam) 2016 Yu, Ye 整理了不同形态学研究中影响城市布局的关键要素 However, while there is a growing body of theory on the subject, collaborative urban spatial generation is still considered to be a difficult process 用不上了 classical approaches are based on very limited observations and inconsistent methods. in the traditional way, designers spend a lot of time working on sketching for different characteristics of urban design in the initial stage for a SUP 换词 Historically, urban morphologist study cities based on qualitative approaches and personal observations and limited to few famous cities or even one specific location1–3.（2017Moosavi, Vahid） morphology 否定 It is difficult for human designers to quantitatively analyse a large number of urban form examples in a short time and quickly to generate planning solutions based on urban contexts how to apply a holistic approach to the large number of urban morphological characters and morphological structure to assist Synergetic urban plan generation in a fast way is a challenge 量化的方式去解决城市转型 电脑辅助量化的城市形态来快速模拟城市转型 Computer-aided a quantitative urban morphological factors provide potential for fast generation and sustainable urban transformation 核心 the static and sectoral approaches cannot fully fulfill such complex and dynamic requirements. computers are playing an increasingly important role in the design process. Man phases in design once carried out by hand are nowadays automated (Bolognini et al. 2011). The city is not made up of independent elements, a synergistic solution sup requires a integrated consideration of the urban system structure rather than a single urban element. Various computational methods in studies of urban morphology have been employed to define and distinguish patterns of a city, such as the statistical model (Colaninno et al. 2011), space syntax (Alexander 1965), convolutional neural networks (Al- bert et al 2017), among others. Much research has been done in different urban sciences with the consideration of the city as a complex system (2011Manesh, Shahrooz Vahabzadeh Tadi, Massimo) TADI 2011 they develop a of construc- tive methods to improve upon their existing environmental performance by TRANSFORMATION OF 形态学form via imm computeral method miao yufan 2018 developed a computational method for rapid urban design prototyping based on an urban designer’s Morphological requirements however Boeing15 argues current quantitative studies suffer from main limitations of small sample sizes, excessive network simplification, difficult reproducibility, and the lack of consistent and easy-to-use research tools. besides ,by only use the of detailed analytical metrics of urban morphology , result of generation sometimes fail to capture contextual and holistic aspects of the urban morphologies . It is important to analyse a large number of non-over-simplified samples holistically to support generation ai 和ml 的应用 compuer assistant for 城市形态学分析 From evaluation to the generation of a sup, ML play an important role 原理图 The computeral automated urban analysis especially based on deeplearning enables the rapid learning of urban characteristics from a large number of prototypes and the rapid processing of complex data messages. This offers the possibility of a holistic approach to the generation of sup devlopment. （Moosavi, Vahid2017）trained a deep convolutional auto-encoder, that automatically learns the hierarchical structures of urban forms at the global scale (2013Hecht, Robert) developed a automatic derivation of urban structures types method by using （ml） with focus on residential areas. 2020Rhee, Jinmo on the analytical potential ofmachine learning methods for urban analysis. he developed a new method for data-driven urban analysis based on diagrammatic images describing each building in a city in relation to its immediate urban context. However, their research focuses on analysing the similarity and clustering characteristics of a large number of urban forms. There is a lack of generative methods to help designers to deal with a large sample of urban morphology and to quickly generate a sustainable urban design. generative aproach with machine learning method The following paper uses computer vision related methods to complete a preliminary investigation in the field of architectural design generation In 2018, Hao Zheng used pix2pixHD to train 100 samples for the recognition and generation of architectural interior design layouts. Stanislas Chaillou has produced the interior layout generation program ArchiGan in 2019. it outputs the interior layout when the site conditions are entered. Human intervention is implemented to generate the results. The schgan produced by Yubo Liu in 2019 can automatically generate functional layouts for primary schools based on road and contour conditions. The results demonstrate that deep learning can enable the generation of general layouts. (2020JIAQI SHEN1, CHUAN LIU2) apply GAN in creating urban design plans, helping designers automatically generate the predicted details of buildings configuration with a given condition of cities. The above articles have the following questions: 1多集中于小尺度建筑平面生成，少有城市尺度的生成方法 2 Due to the limits of algorithm and the number of samples, the generated results have a lot of noise, and the visual recognition is greatly affected; (3) The above researches can not realize diversity generated results with a single input, which is impossible to provide designers with multiple possibilities. 4 没有建设性的方法可以和生成城市形态来改变 goal for generative tool 总结 Summarizing the above literature, it is found that the main difficulty of sup generation is 1 The traditional method requires artificial perception and analysis of a large number of morphological forms, with a small sample size and inaccurate perception 2 It is difficult to use a holistic approach to urban planning to synergise the various stages of the project and the various roles to end this This paper proposes an interactive way to automatically generate urban prototype by deep learning model and morphological rules considering a wide range of urban contexts, rather than a single urban element The synergy between government strategy and the realities faced by citizens reduces the perceived inaccuracy and duplication of work in the traditional way. An interactive tool that takes morphological design rules as input and processes them through a computer-trained database to output image data showing predicted details of integrated urban planning and building configurations. configurations that fulfil all restrictions, which are primarily 1.in concept stage: The tool automatically generates the complet urban street networks and predicts the neighborhood layouts by entering the site boundary ,entrances and street network nodes 2in the mid-term ：The tool can generate urban building layout and detail spatial configurations such as height ,function,density,Mix of functions by input morphological features (site boundaries ,street network openings of boundary ,vacant area ) 3 for the project optimisation stage, vectorising multiple urban solution 3d models with visualazation performance (accessibility functional mix) making it possible to test and evaluate different variants rapidly ## Method ## Urban morphology database for ml Open Street Map (OSM), which is an open source project that provides a free and publicly available map of the whole world. osm data platform Provided data including the morphological characteristics of the buildings and their surrounding urban structure types. To train the networks, the dataset of different milan maps were collected and Morphological characteristics of the map were extracted in an automated way According to the classic city generation process, two city datasets are created for solving the different scales of generation. 1level dataset :For urban street networks generation and neighborhood layouts prediction such as openning border, vacant area, Morphological characteristics 2leve dataset: For building layout and detail spatial configurations such as height ,function,density,Mix of functions Level 1 dataset osm data collection and coding The data set generation steps are divided into 1 data collection 2 data processing and visualization Data collection： houdini automatically acquires Milan's osm vector map data which contains detailed information such as urban street networks, neighborhood layout, vacant area, building layout, function and height, etc. We use the filtering function of houdini to automatically select samples from within the city of Milan. A square of 768m x 768m was used in houdini to select a sample of land parcels of milan road network map, and parcels with >4 parcels were judged as samples. The sample size of the final dataset was 801 ![](https://wwwD.filepicker.io/api/file/qfVOnCpiROK1ooT4Xkzc) ![](https://www.filepicker.io/api/file/SKrtgZdUR6i88xTeMKpu) Data processing and visualization Before generating a computer-recognizable dataset in architecture and urbanism, wedefine the type of dataset as bitmap data of maps are presented as coloured images, with the different colours channel representing the various elements in the map. As shown in the Figure shows A pair of bitmapas obtained from level1 dataset used in this study a： an label bitmap shows morphological rules as input b ：a bitmap shows true urban street networks and neighborhood layouts as output ![](https://www.filepicker.io/api/file/N8Zi7pESdGTSfk9t9kGF) Data processing The sample is automatically drawn in houdini and exported as a bitmap of 512x512px with a scale of 1px = 1.5m. bitmap drawing princeple for a (input)： The red channel represents the shape of the plot indicated by the colour block The green channel represents the entrance and exit of the block drawn with a point of radius 7.5x The blue channel represents the road network nodes of the block drawn with a point of radius 7.5x buildings是咋定义的？？？？ bitmap drawing princeple for b (output)： 如图b The red channel represents a 6m wide road network within the site to separate the plots. The green channel represents openning of boundary Based on the houdini software?? algorithm analyses and visualizes the openning segment of the site boundary. 如图 If the algorithm detects that there are no obstructions from buildings within a depth of 20m from the point on the boundary to the interior of the site ,then it determines that this point is open and connects the open points to form the open line segment. 20m is the value recommended by the professional architects in Milan and can also be modified according to other cities. ｉｎｔｅｒｓｅｃｔｉｏｎ　垂直　 在边界上采样点向内发射ray撞击建筑边界，撞击点到发射起始点的距离超过20m即open点，20m以内即voild。删除voild部分，得到open小短线再次向内offset 6m，使相邻每个地块的open不重叠。 Blue channel represents the vacant area of the site 如图 Based on the houdini software 2-D polygon offset algorithm analyses and visualizes the vacant area Blue channel: the remaining part of the junction between the building boundary offset 12m outwards and the site boundary offset 12m inwards i.e. vacant area 表达不行 蓝色通道：建筑边界向外offset 12m，场地边界向内offset12m 的交界的剩余部分即 vacant area 12 and can also be modified according to other cities and Define the requirement of different projects ![](https://www.filepicker.io/api/file/tynpxB2JShimPvEED0oE) Level 2dataset osm data collection and coding Drawing of bitmaps for input： The level 2 model and the level 1 database image drawing rules are basically similar and only the key differences are discussed below Data collection： First, using osm to obtain the height and function data of all milan buildings and to group the functions of the buildings into 7 categories Accommodation,Commercial, Religious, Civic, Agricultural , Facility , others The sample was then filtered to meet the morphological requirements. Each plot was selected if it had mix of building functions > 0.3 (number of building functions / number of buildings) building density > 0.3.*(building area / land area) spacesyntax nach r500m< 2.5 (number of building functions / number of buildings) building density > 0.3.*(building area / land area) A total of 1017 samples were selected as dataset ![](https://www.filepicker.io/api/file/QWbNfI9S1V7SoYyn1fCg) data Data processing and visualization The sample is automatically drawn in houdini and exported as a bitmap of 512x512px with a scale of 1px = 1.5m. ![](https://www.filepicker.io/api/file/k9CnjD7RRCOc3HnQygui) As shown in the Figure shows A pair of bitmapas obtained from level2 dataset a： an label bitmap shows morphological rules b ：a bitmap shows true building layout and detail spatial configurations Drawing of bitmaps for a ： The red channel represents the level2 plot shape indicated by the colour block The green channel represents openning of boundary(参考level1 dataset ) Blue channel represents the vacant area of the site(参考level1 dataset ) Merge multiple bitmap channels to obtain Figure a Merge multiple color channels to obtain bitmap a in Figure x Drawing of bitmaps for b： The red channel represents the separation line between buildings with a width of 2px Green channel represents building height information, G = clamp(building height / 25.0, low=0, height=1) * 0.8 + 0.2) The Blue channel represents the functional information of the building. Different building functions represent different values of the blue channel B：Accommodation = 0/6, Commercial = 1/6, Religious = 2/6, Civic = 3/6, Agricultural = 4/6, Facility = 5/6, others = 6/6 Merge multiple color channels to obtain bitmap B in Figure x ## generation of GAN machine learning modular GauGAN GauGAN neural network architecture GauGAN is an image translation algorithm published by Nvidia Lab in 2019 that can achive multi-modal synthesis. This experiment is implemented according to the paper, including VAE, which is used to achieve style guide multi-modal synthesis. The implementation of GauGAN is shown in Fig. 1.4.3(1). uses the GauGAN algorithm to complete the model architecture and us-es step-by-step training (outline-> buildings and buildings-> roadmap) ![](https://www.filepicker.io/api/file/EyMWSPczSYedCYkAG73h) step training. step training is to map the contour graph to the building layout graph first, and then map the build-ing layout graph to the road network graph. When testing, matrix multiplication is applied on the obtained building layout and road network to obtain the final re-sult. The training process is also divided into two steps. 1 level traning model 2level traning model ![](https://www.filepicker.io/api/file/jlUazpqcRReSICIgSpRK) ![](https://www.filepicker.io/api/file/0inNDIY0T7Oo3jrupSHY) 1080ti pytorch， 两个模型都用了21小时， level1是801个样本，level2是1017个样本（给于老师） 训练出来的link test moudular test 流程流程： 1.输入大地块边界与出入口并转化为512x512px的位图 2.通过一级神经网络模型生成512x512的带有R:路网（用于地块划分），G:open以及B:empty 3.然后分别矢量化（potracelib C++）三个通道为多边形，并将用路网划分出地块，将open以及empty记录到每个地块上。 4.转化每个地块信息为位图（bound，open，empty）并使用二级神经网络模型进行计算生成R:建筑边界,G:建筑高度，B:建筑功能的位图结果 5.分布矢量化建筑边界用于区分每个建筑的区域，用得到的每个建筑的区域在位图结果中计算该区域的平均Green值以及blue值，建筑高度 = (G - 0.2)/0.8 * 25(米)，功能与blue通道值对应关系见二级模型output制作 Vectorization and 3D procedural modeling Models trained by the step training generates The bitmap results are vectorized and merged, and then 3d procedural modeling and visualization are performed. 如图 经过两个ａｉ　模型训练之后，分别得到路网和building　ｌａｙｏｕｔ　的位图。 两个位图result 的着RGB 三个颜色通道分别对应着不同的建筑讯息，在houdini中利用 potracelibC++ 将 一级模型RGB和二级模型的Ｒ通道矢量化之后就会得到2d的矢量地图。二级模型ｒｅｓｕｌｔ的Ｇ通道记录了建筑高度讯息，Ｂ通道记录了功能讯息。 建筑高度 = 反函数 G = clamp(building height / 25.0, low=0, height=1) * 0.8 + 0.2) 建筑功能　B通道：Accommodation = 0/6, Commercial = 1/6, Religious = 2/6, Civic = 3/6, Agricultural = 4/6, Facility = 5/6, others = 6/6 根据上述的定义如图的几种参数化building　原型，被矢量化的的通道讯息用作参数化建模得到ｕｒｂａｎ　３ｄ　ｍｏｄｅｌ　ｗｉｔｈ　ｄｅｔａｉｌ　．　 使用者能够直观的评估空间关系，和进行性能模拟。 ![](https://www.filepicker.io/api/file/kQoDsmwQLGoil673FkAi) case study The research presented here evolved within the context of The International Design Competition “Attra\verso San Cristoforo”IDCA hold by milan administration in December 2018 The objective of the project is to improve strengthening of the public and sustainable transport system by Regenerate the urban road network and building layout The problem with the current site is that the unreachable transport of disused railway area and Unconnected urban layout lead to low performance with functional diversity and density as well as low accessibility of public space. The difficulty with urban regneration is that designers need to analyse a large number of case studies of the relationship between complex urban forms and corresponding performance , and design strategies based on perceptions. This approach has a small sample size and is not accurate. We worked on this project in collaboration with the archiford architecture company.Our main task is to provide designers with an ai tools for Automatically learnning a large number of samples and generating urban design by deep learning model and morphological rules considering a wide range of urban contexts ![](https://www.filepicker.io/api/file/CBDYEIM6Qi6xcP9a9PZe) reqiurement The requirements from IDCA and the designers can be summarized as follows: the tool should allow to make efficient use of the available space on the site by improving the building density and function diversity performance, but at the same time ensuring their high accessibility to public space performance. morever,street network Needs to be modified with better connectivity and increased open segments that serve the surrounding districts ![](https://www.filepicker.io/api/file/luddh9UITb6hgCy8gBj6) 如图展示了URBANDESIGN的生成的流程 ![](https://www.filepicker.io/api/file/MNaj3eb2RFO5SwgJ5aiu) 1 Generation of Street Networks and nabourhod layout ![](https://www.filepicker.io/api/file/LNEGraLRkCNlug2YSowW) The original road network of the base is not well connected to both sides of the river and there are few access points to the centre of the base, resulting in a low accessibility of the site. It is difficult for residents to reach the public spaces in the middle of the city because the lack of connectivity on both sides / We need to increase the number of entrances to the original base in green, and key road nodes on both sides of the network in blue.　 For the generation , inputing and ajusting Location and number of entrances Location of road network nodes A bitmap is obtained and fed into the ai model to obtain a quick solution ![](https://www.filepicker.io/api/file/EWz3ovT6CYKWMR3wSfFA) Several parameters are introduced to measure the performance of the AI design as follows vacant land = sum(blue areas)（就是empty面积） open border = sum(green lines)（就是open总长度） accessibility = sum(node_count_r400m ^ 1.2 / (total_depth_r400m + 2))（每条道路的，空间句法路网活跃度Nach_r400m，之和） 2012　Hillier, B.; Yang 上述结果的形态显示出不同的路网链接分布。大多路网都能被路网入口和交叉点链接展示了输入因子对于方案的控制能力，但是少数方案的路网会不连续如方案4和8.5号方案的南北两侧连接性最强，opening of boundry 长度也很大所以选择了五号案例作为方案。 ![](https://www.filepicker.io/api/file/uE18vxMTxKWGUweSLYLq) ａｉ　ｍｏｄｅｌ　输出结果如图ｂ表示的　ｓｔｒｅｅｔ　network showing openｎｉｎｇ　ｓｅｇｅｍｅｎｔ　ｏｆ　ｂｏｕｎｄａｒｙ　ａｎｄ　ｖａｃａｎｔ　ｌａｎｄ　 和原基地对比，新的路网有更好的河岸两侧连接性和更多边界和基地核心的通道，绿色通道表示的开放性比片段让整个基地的渗透性更强，空地的分布更加零散但却是不规则的/ 参数显示　和原来的路网比较　ａｉ的生成的方案拥有更好可达性，更多的开放边界，更多的空地。基本满足我们的项目需求 Generation of buildings ｌａｙ　ｏｕｔ　and Space configuration For the generation , 我们根据１ｌｅｖｅｌ　生成得到ｂｉｔｍａｐ图　，输入到ａｉ模型中即可快速的得到building　ｌａｙｏｕｔ。　也可以人工调整该方案的　 ｏｐｅｎ　ｂｏｒｄｅｒ　 ｖａｃａｎｔ　ｌａｎｄ　 street　ｎｅｔｗｏｒｋ　参数等 we generate 不同的ｂｕｉｌｄｉｎｇ　layouts as depicted in figure xx ｂｙ　ａｉ　ｌｅｖｅｌ２　ｍｏｄｅｌ　 经过输入的参数调整，得到多种方案如下。 ![](https://www.filepicker.io/api/file/hSQF5QIMR1aFmXHT0REF) 这里我们引入了参数来衡量ａｉ设计的performance building density = 建筑面积 / 地块面积 function diversity = 每个地块的，建筑功能数 / 建筑个数，的平均 average height = 建筑平均高度 accessibility = 每个建筑到vacant land的最短距离，的平均值 从布局中我们也能发现布局拥有独特的米兰城市空间形态，基本符合米兰的形态规律　。建筑以院落形式沿建筑路网分布提升了空间的密度和空间效率。open 很好的影响了建筑的边界开放， 但是ｖａｃａｎｔ　空间较少的时候方案的建筑界限难以分辨，需要人工简化如ｃａｓＥ４ 参考performance和形式　可以看出building的形式随着不同的输入细节的变化，其中７方案和原来方案的对比能表现出更高的可达性的同时　建筑的密度和功能混合也增加。我们选择7方案作为建筑layout plan ![](https://www.filepicker.io/api/file/bLv67qJeSlqUWy3FBJzp) 根据输出的位图矢量化（method的３章）得到完整的３ｄｌａｙｏｕｔ方案ｓｈｏｗｉｎｇ　建筑具体讯息　高度　功能分布　等等，矢量的模型可以用于模拟和方案绩效研究给建筑师和非专业者提供了很好的协同性平台。 ![](https://www.filepicker.io/api/file/y2urMnMrQ4CaTucxVPfp) ![](https://www.filepicker.io/api/file/ZUfAUcHvTJurYcDMnlZJ) 参数化代理了更加细致的模型，这有助于建筑师在概念阶段就能通过ｔｏｏｌ　预览最后的方案细节和功能划分，和绩效。有助于建筑师和规划人员做规划决策 ![](https://www.filepicker.io/api/file/SbYc0UnZQtGocAqnAZlR) 我写到这目前　　。。。后面编编估计可以了就 ##result Discussion With respect to the aims outlined in the introduction, the results presented in the case study show that we were able to fullfill the requirement by our ai model in a short time . The results of the generative methods for the case study area made it possible to automatically generate urban layouts based on 形态学元素 These layouts also satisfy the requirement to create high density and function diversity performance building layout , but at the same time ensuring their high accessibility . Moreover, an urban designer can interactively revise the urban layout by moving street segments to achieve more satisfactory results. 2.advantage 1 The results of the generative methods for the case study area made it possible to automatically generate urban layouts based on 形态学元素 in a interative way . 传统依靠人类感知和案例分析的urban design 的流程 被ai model 优化缩短到数秒钟 2These building layouts also satisfy the requirement to create high density and function diversity performance building layout , but at the same time ensuring their high accessibility . street network performance of result also was improved in connectivity and openning segments that serve the surrounding districts 4 特定形态而不是随机的urban form 的样本识别和机器学习，提高了performance的同时, 形态上 更趋符合城市语境，近于设计师选取的意向 5 3future work we also discovered that our method has some inherent limitations and did not fully address the needs of urban designers in practice. ｖａｃａｎｔ　空间较少的时候方案的建筑界限难以分辨，无法作为建筑方案如case4 需要人工简化.vacant 的分布和形态对建筑布局影响很大 2位图和矢量图转换和ａｉ之间的协调性 还不是一体化的，之后可以集成在一个平台 3在level1中 少数方案的路网会不连续如方案4和8.不连续的路网虽然可以被人工修改但是影响了level1和2 模型的协调 目前只能在houdini平台处理建筑图形数据传输到gaugan 模型计算，这个过程是耗时的并且需要依托于houdini 平台的矢量转换，我们将要整合所有的功能到web 端为了之后更简单的协同工作流程 并且我们考虑将两个解决不同尺度问题的gaugan模型整合成一个能够同时解决多尺度生成的gaugan模型这需要很大的算力 Currently we can only work on the houdini platform to transmit the architectural graphics data to the gaugan calculation, which is a time consuming and dependent process. acknowlagedment 政府项目 ai archiford.com米兰运河项目规划方案就是3个设计师五天完成了这个项目 ，用这个项目做的 ,并且进行了相关项目投递，节约了时间 . 机器学习的模型在这 ## Conclusion ## References ## List Or you can simply list your references here: 1. some ref 1. some other ref. Numbering fixes itself automatically. 2. A third ref. # References 1. Xuan Yang (2011) The Effect of Neighborhood Mode in Modern Residential District Plan-ning. Chinese & Oversea Architecture, 94-95 2. Jia Rongxiang, Sun Ying (2012) Cultural Characteristics and Existence Value of Buildings in Beijing Million Village. Journal of Beijing Institute of Civil Engineer and Architecture, 76-80 3. Li Hao (2014) The introduction of Soviet planning theory to China. Beijing Planning and Construction, 165-168 4. Dan Hendrycks, Kevin Gimpel (2016) GAUSSIAN ERROR LINEAR UNITS (GELUS), University of California, Berkeley, Toyota Technological Institute at Chicago https://arxiv.org/abs/1606.08415 5. Weixin Huang, Hao Zheng (2018) Architectural Drawings Recognition and Generation through Machine Learning, Tsinghua University, University of Pennsylvania https://www.researchgate.net/publication/328280126 6. Stanislas Chaillou (2019) AI Architecture Towards a New Approach https://www.academia.edu/39599650/AI_Architecture_Towards_a_New_Approach 7. Phillip Isola, Jun-Yan Zhu, Tinghui Zhou Alexei A. Efros (2016) Image-to-Image Transla-tion with Conditional Adversarial Networks, Berkeley AI Research (BAIR) Laboratory, UC Berkeley https://arxiv.org/abs/1611.07004 8. Ting-Chun Wang, Ming-Yu Liu, Jun-Yan Zhu, Andrew Tao, Jan Kautz, Bryan Catanzaro (2017) High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs https://arxiv.org/abs/1711.11585v1 9. Taesung Park, Ming-Yu Liu, Ting-Chun Wang, Jun-Yan Zhu (2019) Semantic Image Syn-thesis with Spatially-Adaptive Normalization https://arxiv.org/abs/1903.07291 10. Yubo Liu, Qiaoming Deng, Linyu Liang (2020) SchGAN：Primary school campus layout generation, intelligent assistant for architectural design https://blog.csdn.net/shadowcz007/article/details/104035601 15. Boeing, G. Osmnx: New methods for acquiring, constructing, analyzing, and visualizing complex street networks. Comput. Environ. Urban Syst. 65, 126–139 (2017). casestudy urban transfortmation of naviligio grande casestudy the urban transformation of naviligo grande 让机器像罗西一样思考