以GMDH演算法发展铺面评价系统-科技部工程科技推展中心.PDF

行 精 類 行 年年 行 參理 理 劉 論 理 年 摘要 GMDH （Group Method of Data Handling ）演算法為俄國學者Ivakhnenko 於 1967 年提 出的多層次輸入－輸出關係之自組性網路架構，具備毋須任何統計假設即可由實際數據資 料自行推演組合各種輸入變數間之關係的特性，本研究將使用 GMDH發展鋪面養護與維修 （Maintenance and Rehabilitation, MR ）決策模型。傳統常透過人為目視記錄各種鋪面破損 之範圍（coverage ）與嚴重程度（severity ），並主觀判斷所應採行之MR 策略，由於導致 破損的因素繁多且交互影響複雜， MR 策略的決定又存在複雜的主觀感認變異，故 MR 決策模型的發展符合 GMDH的理論機制。本研究採用由養路工程師自台灣北區縣道收集之 多種鋪面破損類型，及各路段相對應採行之MR 策略共計 2,348筆實際紀錄（ 2,115筆供 模型發展、 233筆供模型測試） ，以18種破損類型為輸入變數、 MR 策略為輸出變數，發 展完成 MR 決策模型，依據測試紀錄驗證模型的總正確率達0.940，由此可預期基於GMDH 所發展之決策模型可降低人為主觀判斷 MR的變異，協助決策過程。 關鍵字 : GMDH ，鋪面破損，養護與維修 ABSTRACT Group Method of Data Handling (GMDH), was proposed by Ivakhnenko in 1967, is based on the principles of heuristic self-organization that is developed to solve complex problems, such as pavement maintenance and rehabilitation (MR) decisions. In the paper, an empirical study is conducted to develop MR decision model using GMDH. Pavement distress surveys were performed on seven county roads by experienced pavement engineers. For each road section, the severity and coverage of existing distresses and required MR treatment were separately recorded. The analytical database consisting of 2,348 records (2,115 records for model development and 233 records for model testing) are established to develop and test the MR decision model. The finding shows that the total accuracy of proposed model is 0.940 based on the testing records. It illustrates that the GMDH-based MR decision model certainly can assist en