一個基於BIM與GIS的不動產估價方法

Abstract

以往建築公司在決定建案或設計圖之前，會經過一定的評估程序。其中，會對於建案或設計圖做價格評估，以利財務估算。然而，透過人工來估價是費人且費時的。因此，國內外開始有研究針對此估價的不便來提出解決辦法。在眾多的資訊技術當中，以人工智慧為基準的方法，例如類神經網路(ANN)，被採用的最為廣泛。原因在於，其可以模擬非線性系統，非常符合不動產估價的不確定性。採用類神經預測的研究中，有三個問題存在：1.著重在房屋不動產2.轉換資料時，為作者的主觀想法3.採用人工輸入數據。本研究提出一套價格評估方法，來解決這些問題。首先，結合建築資訊建模(BIM)與地理資訊系統(GIS)解決人工輸入數據的問題。再透過模糊理論與共識理論取得客觀的資料轉換方法。且透過建立分類與多個類神經網路模型而能預測不同種類的不動產標的。在實作系統上，分成兩階段。第一階段以訓練類神經網路模型為目的，透過實作工具MATLAB與台灣政府實價登錄的資料為依據，所訓練出來的模型誤差率大約在正負6%之間，證明所提出的方法論與模型具有高度的正確性與可靠性。第二階段以建立自動化系統為目的，透過JAVA程式，將第一階段訓練出來的模型與建築資訊建模、地理資訊系統的資料結構做結合，開發出完全自動化的系統。使用者將建築資訊建模與地理資訊系統的圖檔放入自動化程式，即可立即評估出該建物的價格，以提供決策者決策參考，節省人力與時間消耗。

The traditional real estate appraisal and price prediction are inefficient and expensive. Therefore, many of the researches use information technology to solve these problems. Among the information techniques, the artificial neural network (ANN) that can simulate complicated nonlinear system is adopted frequently. However, related researches which use ANN to predict the price have three problems: limited only to residential buildings, subjective data transformation, and manual data input. Therefore, this thesis proposes a prediction method to solve these problems. First, it integrates the Building Information Modeling (BIM) and the Geographic Information System (GIS) to solve manual data input problem. Then, the fuzzy set theory and consensus are used to reduce bias on data transformation, and classification techniques are used to predict a variety of real estates in addition to residential buildings. As to system implementation, there are two phases. Phase one aims at training the ANN models. Both MATLAB and open data from the government of Taiwan are used to train the models, and the subsequent error rates are close to 6%. This result indicates the proposed method is highly accurate and reliable. Phase two aims at system automation. A JAVA program is used to integrate both BIM and GIS data with the trained ANN models. A user selects both BIM and GIS files as inputs to the system. The system automatically gives decision makers a predicted price.