支持复杂规范条文解译的基元函数提取与分析

ATOMIC FUNCTION EXTRACTION AND ANALYSIS FOR INTERPRETING COMPLEX RULES FROM BUILDING CODES

  • 摘要: 规则条文解译(rule interpretation)是智能审图(automated rule checking, ARC)面临的关键挑战之一,目前依赖领域专家完成规则解译,耗时耗力。其根本原因是,规范条文包含大量复杂逻辑与领域隐含知识,计算机缺乏相应的基本常识与逻辑规则,因此难以有效理解和解译复杂设计规范条文。因此,该研究以《建筑设计防火规范》关键内容为例,通过系统整理、分析与拆解有关规范条文,首次提出了基元函数概念,以表征不同规范条文规则推理与逻辑计算过程共用的基础算法单元;并面向复杂规范条文解析,建立了有关基元函数的识别与提取方法,识别、定义了存在性检查、数量检查、几何检查、距离检查、面积检查、空间位置检查、窗墙比检查、属性检查共8类66个基元函数,构建了首个针对我国建筑结构设计审查的基元函数库。同时,研究从审查对象以及基元函数复杂度、使用频次、共现特征与复杂条文表示能力等方面进行了详细分析。结果表明:基元函数使用频率呈长尾分布,属性类函数与其他各类函数都有较高的共现次数,所建立的基元函数具有强大的复杂逻辑表示能力,可大幅提升复杂条文解译范围与推理能力。

     

    Abstract: Automated rule checking (ARC) is revolutionizing in various domains, while its key stage, rule interpretation, still remains a bottleneck. Currently, the process relies heavily on domain experts, making it both time-consuming and labor-intensive. This challenge arises due to the complex computational logics and hidden domain knowledge used in regulatory clauses. These complexities are compounded by the computer's inability to inherently understand complex design rules, given its lack of basic common sense and logical rules. Therefore, the key contents of "Building Design Fire Code" are taken as examples, and through systematically organizing, analyzing, and disassembling the relevant rules, the concept of "atomic function" is firstly proposed. This concept aims to characterize the basic algorithmic units shared by different rules when it comes to rule reasoning and logical computation. Then, towards the analysis of complex rules, the method to identify and to extract relevant atomic functions is established. As a result, 66 atomic functions are identified, and further categorized into 8 groups for the checking of existence, quantity, geometry, distance, area, spatial location, window-wall ratio, and property respectively, thereby forming the first atomic function database for the checking of Chinese building design codes. The complexity, utilization frequency, co-occurrence patterns, and complex logic representation capabilities of the checking objects and atomic functions are analyzed in detail. Results show that: atomic functions' usage showcases a long-tailed distribution; property retrieval functions frequently co-occur with other function types; the atomic functions established have a strong capacity for representing complex logics, which can greatly improve the interpretation scope and reasoning ability of complex design rules.

     

/

返回文章
返回