伴隨經濟與都市之發展，民眾生活品質日益提高，市區道路管理之意識高漲，道路施工品質隨著技術層次提升進而改善使道路服務水準，促使市區道路路網發展成形，各政府機關須有效利用資源以管理道路並施行養護作業。依據內政部營建署針對市區道路所進行之「市區道路養護管理暨人行環境無障礙考評計畫」，考評結果得知各縣市政府於市區道路養護管理上逐漸進步，但尚有部分問題需要改善，因此，本研究欲透過市區道路服務品質及道路養護經費進行關聯性分析，探討如何於有限經費下達到一定水準之道路服務品質與平坦度改善效率。另外，將各縣市之道路相關因子進行比較及各別性分析，深入了解各縣市各年度之基本資訊及相互關係，並透過三種人工智慧統計分析方法，分別為遺傳表達規劃法(GEP)、廣義迴歸類神經網路(GRNN)及支持向量迴歸(SVR)，建構適用於各縣市政府未來考評計畫之分數預測模型，經本研究比較分析後，SVR分析結果為最佳，其判斷係數R2達0.9表示具高度解釋能力，故採用SVR建立預測模型之程式。以臺北市做為個案研究，得知當養護經費提高道路服務品質隨之提升，並將所得之估算經費帶入建立之模型中，最後綜合探討適用於臺灣市區道路養護管理之策略。;Urban road service level is always concern by resident in Taiwan area. According to the survey of “Urban Road Maintenance Management and Pedestrian Accessible Environment Assessment Project” which was held by Construction and Planning Agency, Ministry of the Interior (CPAMI), local governments have made progress in maintenance management of the urban road. However, there still are a few problems needing to be solved. The research aims to obtain correlation analysis between urban road service quality and funds of road maintenance, and discuss how to improve the performance of road service and the road roughness level under limited funds. In order to deeply explore the basic information and interrelationships of each city and county in Taiwan. This research compared and conducted an analysis for each parameters of urban roads in Taiwan, and constructed a prediction model for future assessment project of governments using three Artificial Intelligence (AI) analysis methods included Gene Expression Programming (GEP), General Regression Neural Network (GRNN) and Support Vector Regression (SVR). After completing the comparative analysis of the research, SVR analysis showed the best performance in forecasting of the three methods. The coefficient of determination (R2) is 0.9 in a high level of interpretation, indicating that SVR model is good enough to establish the best prediction model of the project. The most important parameter would influence the roughness is funding. In the case study of Taipei City, the quality of pavement performance will significance be raised by the increasing of the maintenance funding, and the SVR prediction model could be utilized in field. Based on above, this study is worthy of further study in the future.
