適應性類神經模糊推論系統預測公共工程決標金額之研究－以高雄市政府水利局標案為例

APPLYING ANFIS IN PREDICTING THE FINAL BID PRICE OF PUBLIC CONSTRUCTION PROJECTS- CASE STUDY ON KCG WATER BUREAU PROJECTS

王建文(Jian-Wen Wang) ； 王裕仁(Yu-Ren Wang)

《營建管理季刊》 96期 (2013/11) Pp. 25-41

https://doi.org/10.6505/CMJ-2013-96-25-17

摘要

國內現行公共工程投標程序都設有投標期限，並採用最低標方式決標來決定執行工程專案之營造廠商，這不僅迫使營造廠商必需以快速且精準的去推估投標金額，且易陷入惡性削價搶標的激烈競爭，進而影響營造廠商的利潤及得標之效益。故此，本研究將建構一套較精確投標決策作業前的標價預測模式，是多年來許多營造廠商想探討之議題。本研究擷取高雄市政府水利局近2年360筆之工程決標案例資料，得知大部分標案集中於100萬至5千萬元之間，占總案例的86%，故本研究將100萬至5千萬元之間決標案例區分為三個級距來預測分析，並以預算金額、押標金、履約期限等三項因子視為自變數，而工程決標金額設定為應變數，以複迴歸分析、支援向量機及適應性類神經模糊推論系統等方法，來建構模型運算各自變數與應變數之間的關聯強度，並比較不同方法所建構模型之預測差異性。研究結果顯示，在100萬至200萬、200萬至400萬及400萬至5000萬這三個級距，傳統統計方法(複迴歸分析)劣於人工智慧方法(支援向量機與適應性類神經模糊推論系統)。其中複迴歸分析平均相對誤差百分比為83.50%、26.37%及82.62%；而在人工智慧方法之適應性類神經模糊推論系統平均相對誤差百分比為17.25%、15.78%及14.55%略優於支援向量迴歸的18.16%、16.86%及14.9%。較佳的預測模型可作為日後營造廠商是否參與競標及投標之參考依據。

關鍵字

工程決標價 ； 預算金額 ； 押標金 ； 履約期限 ； 複迴歸分析 ； 支援向量機 ； 適應性類神經模糊推論系統

並列摘要

In the public construction project tendering process, there is limited time between RFB and submittal, and the contracts are always awarded to the lowest bid. The current tendering practices force the contractors to prepare the bid in a short time frame, thus the estimate might not be accurate and it will affect the contractor’s profibility. As a result, this research proposes artificial intelligence based bid price prediction models to assist the contractors with the bid price estimation. This research collects tendering project information from Kaohsiung City Government Water Bureau for the past two years. Information from a total of 360 projects is collected and 86% of the projects are within the range of 1 million and 50 million NTD. These projects are divided into three project size groups and three prediction models are developed separately. Budget amount, bid bond and contract duration are set as independent variables in the model and final bid price is set as the dependent variable. Multiple regression is first applied to analyze the data and followed by artificial intelligence techniques (Supprot Vector Machines and Adaptive Neural -Fuzzy Interence System). The prediction abilities between these models are also compared. The results show the in all three project size groups (1M~2M, 2M~4M, and 4M~50M), multiple regression analysis does not yield good prediction results when compares to artiticial intelligence models. The mean error percentages are 83.5%, 26.37%, and 82.62% for multiple regression prediction models for the thress project size groups respectively. The mean error percentages are 17.25%, 15.78%, and 14.55% for the ANFIS prediction model. The mean error percenteges are 18.16%, 16.86%, and 14.9% for the SVM prediction model. The ANFIS prediction model yields the best results and can provide valuable information for the contractors in the bidding process.

並列關鍵字

Project Bid Price ； Budget Amount ； Bid Bond ； Contract Duration ； Multiple Regression Analysis ； Support Vector Machine ； Adaptive Neural Fuzzy Interence System