隨著台灣的土地開發及發展,集合住宅已成為各都會區主要的居住型式。「住」為人類生活的基本需求,相較於其他類型的建築,住宅空間幾乎是每個家庭皆需要擁有的,因此集合住宅也成為許多建築師事務所主要的業務項目。而集合住宅格局與工程造價亦會反映在商品價格上,因此本研究期望能探討集合住宅格局設計與造價間的關係,以便從中瞭解較適當的設計模式。 本研究蒐集台灣地區各時期集合住宅平面,經由專家評分後建立資料庫,透過資料庫訓練類神經網路學習專家的知識與經驗,以建立類神經網路推論模型。然而類神經網路雖能準確推論,但無法得知其運算過程中各因子的權重,此時即可藉由灰色理論進行分析,透過灰關聯係數的計算,探討各因子對於集合住宅格局性價比的影響。 由於集合住宅性價比之因子種類繁多,不易評估其對性價比之影響,故利用主成份分析法將47項影響因子歸納為13項主成份,並引用灰色理論之原理,探討不同戶型(二房、三房、四房)之集合住宅格局主成份與參考序列的灰關聯度。從研究結果可得知,各戶型格局主成份與性價比之灰關聯排序不盡相同。若以二房、三房、四房之總平均而言,集合住宅格局性價比與公共及私密空間組成較相關,其次依序為主臥室、空間面積。 本研究亦運用倒傳遞類神經網路與多元迴歸,推論集合住宅二房、三房、四房之格局評分、單位造價及性價比。從推論結果得知,類神經網路較多元迴歸更適合運用於推論集合住宅之格局評分、單位造價及性價比。研究成果可供設計者參考,研究方法亦可供後續學術研究參考應用。
Along with the development of landscape in Taiwan, condominium has become the major composes of large cities. Being the fundamental element of humanity, “living” is essential for every family, which condominium projects has taken the majority tasks of most architect agency. Furthermore, the structure and construction expenditure of condominium would also reflects the value to the item. This research has conducted the discussion between the condominium designs and costs of constructions, eventually aimed to conclude the most appropriate approach. This research collects several condominium plans with different generations, categorized in to data base after reviewed by specialists, establishing the model of artificial neural networks theories that obtained the specialists’ expertise. Although the accuracy of artificial neural networks can be inference hypothetically, the components of each calculations can not be obtained, which the gray theory can be used in order to explore how the components bring the impacts to the price-performance ratio of condominium. As the variety of price-performance ratio is difficult to evaluate, principal components analysis has been taken to distinguish 47 elements into 13 major components, reference to the grey theory and discussing different type of condominium (with 2 bedrooms, 3 bedrooms, 4 bedrooms). The results had revealed the condominium with different number of bedroom has randomly sequenced. In average, price-performance ratio has more relates to public/private space then the main bedroom, and size of the living area comes at last. This research also applies back-propagation network and multiple regression analysis, sorting the design of 2, 3, 4 bedrooms of condominium in to value, costs and price-performance ratio. The conclusion can be defined artificial neural network is more applicable to evaluate condominium instead of multiple regression analysis. The research results can be used by designers for reference, and the research methods are also can provide academics for references.