基于Octopus和SVMs的住区强排多目标优化设计研究  

Research on Multi-objective Optimization Design of Forced Layout in Residential Area Based on Octopus and SVMs

作　　者：刘德利[1] 王科奇(指导)[1] 闫海燕[2] LIU De-li;WANG Ke-qi;YAN Hai-yan(School of Architectural and Urban Planing,Jilin Jianzhu University,Changchun 130118,China;School of Architectural and Artistic Design,Henan Polytechnic University,Jiaozuo 454000,Henan,China)

机构地区：[1]吉林建筑大学建筑与规划学院,长春130118 [2]河南理工大学建筑与艺术设计学院,河南焦作454000

出　　处：《建筑节能（中英文）》2022年第11期21-27,共7页Building Energy Efficiency

基　　金：国家重点研发计划资助项目:民用建筑适宜室内环境营造基本理念及技术路径研究(2018YFC0704405)。

摘　　要：人工智能技术的发展为设计师提供了新的分析和设计工具,基于机器学习算法,在住区强排设计中引入新的设计方法,在满足良好日照性能的前提下,增加项目的容积率,从而提升土地的使用效率。日照参数和容积率是住区强排设计中两个重要的指标,目前设计者为获得较高的容积率,多采用基于日照模拟分析结果进行多方案对比的方法,效率不高,而使用基于遗传算法和帕累托最优原理的Octopus求解器,对日照性能和容积率两个参数值进行多目标优化可以提升效率。但Octopus求解器进行多目标优化时需要进行多次迭代,每次迭代过程中需要进行日照仿真模拟,耗时较多,因此引入支持向量机,根据多次日照模拟结果数据训练出一个支持向量回归模型,然后用这个训练出的回归模型替代日照模拟软件,可以缩短每次迭代进程的时间,从而提高多目标优化效率。以两个不同的住区基地为例,验证所提方法的应用效果,结果表明该方法能够适应不同的住区基地,所生成方案的日照性能和容积率也达到较为平衡的最优解,能够满足设计师需求。The development of artificial intelligence technology provides designers with new analysis and design tools. With the machine learning algorithm, there are new design methods in the design of forced layout in residential area, which can increases the plot ratio of the project on the premise of meeting good sunshine performance, so as to improve the land use efficiency. Sunshine parameter and plot ratio are two important parameters in residential area design, and it is inefficient that the designer compare multiple schemes to obtain the maximum plot ratio based on the results of sunshine simulation analysis. If the Octopus is used, based on Genetic Algorithm(GA) and Pareto-Principle, multi-objective optimization is carried out for sunshine performance parameters and plot ratio, efficiency can be improved. Unfortunately, Octopus needs multiple iterations for multi-objective optimization, and sunshine simulation analysis takes some time in each iteration, which means multi-objective optimization is an extremely time-consuming process. However, with Support Vector Machines(SVMs), a support vector regression model will be obtained from the data of multiple sunshine simulation analysis results, and it is more efficient that using the support vector model add to Octopus’s multi-objective optimization process instead of sunshine simulation analysis software. Two different sites are used as examples to verify the application effect of the proposed method, and the results show that the method can be adapted to different sites, and the generated solutions reach a more balanced optimal solution in terms of sunshine performance and plot ratio, which can meet the designers’ requirements.

关 键 词：日照性能 容积率 支持向量机 遗传算法 多目标优化 

分 类 号：TU241.2[建筑科学—建筑设计及理论]