預測營建剩餘土石方流向異常行為之研究

NCUIR > Engineering College > Executive Master of Construction Engineering and Management > Electronic Thesis & Dissertation > Item 987654321/93684

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/93684

Title:	預測營建剩餘土石方流向異常行為之研究
Authors:	陳家暐;Chen, Chia-Wei
Contributors:	土木系營建管理碩士在職專班
Keywords:	營建剩餘土石方;預測;分類;電子聯單系統;土石方管理;construction residual soil;prediction;classification;electronic joint consignment note system;residual soil management
Date:	2023-07-24
Issue Date:	2024-09-19 17:26:36 (UTC+8)
Publisher:	國立中央大學
Abstract:	自2018年至2021年，台灣的建設項目每年平均產生超過3600萬立方公尺的營建剩餘土石方。這可能導致非法棄置案件激增，對環境衛生和公共安全構成嚴重威脅。本研究旨在探索電子聯單系統是否能夠有效預測營建剩餘土石方流向的異常行為。此外，本研究調查了電子聯單作為一種新的管理模式是否能夠提高管理效率，減輕營建剩餘土石方管理上的人力負擔。資料收集對象為新北市2022年的所有電子聯單，共超過35萬筆資料。專家建議從全部14個變數中選取7個作為輸入，包括建照號碼、聯合號碼、數量、行駛時間和異常狀態、出場地點和進場地點。研究採用了5種最受歡迎的分類器進行預測，包括多層感知器（MLP）、支持向量機（SVM）、隨機森林（RF）、自組織映射（SOM）和樹狀自組織映射（TS-SOM），並採用5等分交叉驗證來得出結果並進行比較和分析。研究結果得出，RF具有最佳預測效果達99.84％，建議從以下三個方面改進實踐：(1)提供更詳細的行程時間異常警示，(2)在電子聯單系統中增加更多輸入變數，以及(3)建立使用者反饋機制。;From the year 2018 to 2021, the construction projects in Taiwan have generated an annual average of over 36 million cubic meters of construction residual soil. This could lead to a surge in illegal cases of disposal and pose a severe threat to environmental hygiene and public safety. This research aims to explore whether the electronic joint consignment note system can effectively predict abnormal patterns in the flow of construction residual soil. Additionally, it investigates whether the use of the electronic joint consignment note as a new management model can enhance efficiency in management and reduce the manpower burden on construction residual soil management. The data collection targeted at the entire electronic joint consignment notes in the New Taipei City for the year of 2022, resulting in over 350,000 datasets. The experts suggested 7 variables as inputs out of the entire 14 variables, including Construction Registration Number, Joint Consignment Note Number, Quantity, Travel Time, and Abnormal Status, Exit Site, and Entry Site. Adopting 5 most popular classifiers for prediction including Multilayer Perceptron (MLP), Support Vector Machine (SVM), Random Forest (RF), Self-Organizing Map (SOM), and Tree-Structured Self-Organizing Map (TS-SOM), the study utilizes 5-fold cross validation to yield the results and performs comparison and analysis. The findings conclude that RF has the best prediction at 99.84%, suggesting practitioners with improvements of (1) more detailed travel time abnormal warning, (2) more input variables for the electronic joint consignment note system, and (3) user feedback mechanism.
Appears in Collections:	[Executive Master of Construction Engineering and Management] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	19	View/Open

社群 sharing

Loading...