以類神經網路預估建築工程造價之研究

Abstract

工程造價預估在專案初期佔著相當重要的地位，精確的造價預估可有效地協助投資者做出正確的決策。然而，由於建築專案初期的資訊不完整，對產品的定位也僅有一些概念性的需求，如果能從這些資訊中推估出較為準確的工程造價，將有助於降低投資者的風險。 建築工程專案在預估工程造價時，需仰賴歷史資料及經驗回饋，過去常用的方法有經驗判斷法、因素估價法、統計理論等；近年來由於資訊的發達，利用電腦模擬人類思考模式，而發展出類神經網路演算法，被廣泛運用於各種不同層面的研究，在營建工程上的應用則多用於預測及推估。 本研究應用類神經網路中的「倒傳遞演算法」以建立一個工程造價預估模式。收集業界建築工程造價實際案例，將工程直接成本劃分為十個工程主項，並依專案初期可取得或預測的12筆資訊作為輸入變數，分析檢討與各主項工程相關的輸入變數，透過網路的學習訓練及參數改進修正，進行工程造價的預估。研究結果顯示倒傳遞類神經網路模式可得到快速、精確的預估成果，因此適合作為建築專案初期投資效益的決策評估使用。

During initial phase of construction project, accuracy of cost estimate, which plays greatly important role, can be beneficially crucial along decision-making process among interest bodies. Accurate cost estimate may lower risk provided incomplete scope of information as well as undefined project goal. Cost estimation on construction project relies substantially on experiential base of information, such as experience judgment method, factor estimation method, statistical theory…, etc. Until recent years, neural network algorithms, which represents thinking ways of human beings achieved by computational simulation, has been widely applied on research on extensive fields, including cost estimation on construction project. This thesis established one cost estimate model with application of 'back propagation algorithm' of artificial neural networks. Based on existing construction project cases, from which initial phase twelve informal variables were defined as input along computation whereas the project cost were divided into ten main items. Cost estimate can be fast and accurately achieved facilitated by self-analytical development of artificial neural networks and through amendment of the input variables. In conclusion, the application of artificial neural networks satisfies the accuracy of cost estimation and the assessment of rate of profit during initial phase of construction project.