結合形態生成與建築性能評估之前期建築設計程序之建立

建築設計可以被視為涵蓋因何（What）、為何（Why）以及如何（How）三個工作步驟的解決策略（Problem-Solving）程序。回溯既往的學習經驗，不同階段建築設計的學習重點均聚焦在形式操作而非解決設計問題，而在形式操作過程中，對於形式美學的追尋大過於形式與機能的相互連結。設計的『為何』與『如何』被侷限在形式操作過程的合理性而非具體問題與解決設計策略的相互呼應。同時，由於學習過程中所面對的大多數建築設計操作課題，均有明確的建築機能需求指示，學習者絕少能自行釐清，從『因何』到『為何』、從『疑問』到『問題』的思維。同時，過於強調直觀式的形式美學操作訓練，亦削弱了建築機能需求與建築具體形式之間的相互對應關係。建築形式並非純粹出自於獨立的形式操作過程，它實際上是整體解決策略（Strategy）的具體呈現。因此，在設計發展過程中每一階段的設計決策都是有跡可循的，所有形式均來自於明確目的與手段的相互對應，其中並無任何模稜兩可或猶疑不決之處。遵循此一原則，數位演算形態生成應該被視為通過數位化模式將建築設計解決策略程序中的具體問題轉譯成為各個需求變數與相應的數學模式，並以此為依據推導出形式解決方案，而非僅將其視為數位化的形式操作工具。如何將完整的建築設計解決策略程序轉譯成為可行的數位演算形態生成邏輯的演繹與推論程序，為本研究主要之研究動機所在。本研究旨在建立結合形態生成與建築性能評估之前期建築設計程序。首先參考建築量體形式操作範例，將其轉譯為建築量體形態生成程序，並轉換編程為Grasshopper演算步驟，進行建築量體形態生成之邏輯演繹，藉以確認相關形態的生成控制參數。再藉由建築物理環境Ladybug Tools分析插件，就平均日照輻射量對於建築形態生成之影響進行分析。本研究主要的研究變數包括建築量體形態生成程序與其相關的控制參數，以及環境控制參數三者，主要目標希望推論出--『在環境控制參數最佳化的情形下，形態生成控制參數與生成結果之最佳解為何?』。此一問題屬於多目標最佳化問題（Multi-Objective Optimization Problem），依循基因演算法（Genetic Algorithm），最佳化問題之解為最適應種群的基因編碼。而在演算所得每一代中，通過適應度函式計算得出適應度數值（Fitness Value）對種群內的個體進行評估，並按照適應度高低排序種群個體。本研究通過形態生成控制參數產生各代種群個體的基因編碼，並以環境控制參數定義適應度目標參數。之後採用包含基因演算法與帕雷托最優（Pareto Optimal）之 Wallacei X 分析插件，進行形態生成與建築效能評估之多目標最佳化分析。研究結果顯示，變動程序A—Extrude實體路徑向量序列以及實體路徑截面寬度與高度兩種形態生成控制參數，同時變動程序D—Nest建構線序列、建構線點位參數以及虛空間規模等形態生成控制參數，均會增加建築量體總體積與總表面積，從而減少平均日照輻射量並增加平均陰影量。以 Wallacei X 分析插件針對程序A—Extrude與程序D—Nest進行最佳化分析後發現，採用平均適應度級別（Average of Fitness Ranks）分析方法進行最優方案選擇，程序A—Extrude最優方案計算所得之平均適應度級別，趨近於邊界量體與生成建築量體體積差值。而程序D—Nest最優方案計算所得之平均適應度級別，趨近於最終建築量體方案之總表面積。

關鍵字

形態生成；多目標最佳化；基因演算法；基因編碼；適應度目標參數

並列摘要

Architectural design can be viewed as a problem-solving procedure including the three working steps of what, why and how. Looking back on the past learning experiences of architectural design, each stage focuses on formal operation rather than design problem-solving. In the process of formal operation, the formal aesthetics is more important than the connection between form and function. "why" and "how" of the design process are limited to the rationality of formal operations rather than the correspondence between specific problems and solutions. Because most of the building design operation topics have clear functional requirements, during the study. Students are rarely able to clarify by themselves, from "what" to "why", from "question" to "problem". At the same time, too much emphasis on intuitive formal aesthetic operation training also weakens the correspondence between architectural functional requirements and specific architectural forms. Architectural form is not purely derived from an independent formal manipulation process, it is a concrete presentation of an overall solution. Therefore, the design decisions at each stage of the design development process are traceable, and all forms came from the mutual correspondence of clear ends and means, and there is no ambiguity or hesitation. Following this principle, the algorithm generation should be regarded as translating specific problems in architectural design problem-solving procedures into various demand variables and corresponding mathematical models through digital models and deriving formal solutions based on them, rather than treating it only as a manipulation tool in digital form. How to translate the complete architectural design problem-solving program into a feasible digital algorithm generating logic deduction and inference program is the main motivation of this research. The purpose of this study is to establish a pre-architectural design procedure that combines form generation and building performance assessment. Firstly, referring to the operating conventions of building mass, it is translated into a building mass generation program, and then converted and compiled into Grasshopper algorithmic steps to perform logical deduction of building mass generation, to confirm the relevant type generation control parameters. Secondly, the Ladybug Tools building physical environment analysis plug-in was introduced to analyze the influence of the average solar radiation on the generation of building forms. The main research variables in this study include the building mass generation program, its related control parameters, and environmental control parameters. What is the best solution to generate results?". This problem is related to the multi-objective optimization problem. According to the Genetic Algorithm, the solution to the optimization problem is the genetic code of the optimal population. In each generation obtained by the calculation, the fitness value is calculated by the fitness function to evaluate the individuals in the population, and the population individuals are sorted according to their fitness. The research results show that changing program A-Extrude solid path-vector sequence and the two types of solid path section width and height generate control parameters while changing program D-Nest construction line sequence, construction line point parameters, and virtual space scale and other types generate control parameters that increase the total volume and surface area of the building mass, thereby reducing the average solar radiation and increasing the average shading. The Wallacei X analysis plug-in was used to optimize the program A-Extrude and the program D-Nest. The average fitness level (Average of Fitness Ranks) analysis method was used to select the optimal program. The average fitness level calculated by the program A-Extrude optimal solution is close to the volume difference between the boundary volume and the generated building volume. The average fitness level calculated by the program D-Nest optimal solution is close to the total surface area of the final building massing solution.