鋪面服務力指標構建方法之研究

Abstract

目前台灣地區的公路鋪面設計較多採用AASHTO的設計方法進行，輔助以工程經驗再對設計內容及相關參數加以修正。AASHTO設計方法中有許多參數的決定是因地制宜，目前台灣的相關研究也開始針對AASHTO設計方法進行台灣地區適用性之研究，但結果未能普遍實用。因而本研究是以AASHTO設計公式中之現況服務力指標（Present Serviceability Index）之方法為基礎，應用其他關係模式之方法，嘗試發展適用於台灣地區鋪面服務力指標之方法。 本研究中應用模糊迴歸、模糊GMDH及倒傳遞類神經網路三種新的理論方法於鋪面現況服務力指標模式之建構，並且與傳統多元迴歸所建構之鋪面現況服務力指標模式比較，於實證研究結果發現此三種新的理論方法在服務力指標模式的分析應用比傳統多元迴歸優。且經過與工程師討論對於不同模式的實務性，了解明確值對於工程師的決策上之意義並不大，經過比較與討論，在短期少量的資料蒐集上，模糊迴歸較為合適，若能長期蒐集與調查之大量資料，以類神經網路為一較佳的選擇。 模糊迴歸於本研究中應用二次模糊迴歸於模式的建構，由於二次模糊迴歸包含必然性與可能性模式的屬性，於研究中對此項屬性與多元迴歸作驗證，在本研究中結果得知，二次模糊迴歸可得到與傳統多元迴歸相同的結果，且可更能彈性的運用。

Presently, the pavement in Taiwan is almost designed by AASHTO design method which contents and parameters are modified by engineering experience. Many parameters of AASHTO design method take actions that was suited in local circumstance. Presently, other researches also use the AASHTO design method to make the research about the usability in Taiwan, but the result cannot be used widely. This research is based on Present Serviceability Index(PSI) and applies other approaches to develop a new set of approach about PSI which is suitable in Taiwan. In this research, we apply the fuzzy regression, fuzzy GMDH, and back propagation neural network to construct the PSI and compare it with the result which is constructed by traditional multi-regression. It's found that the analysis and application of these three theories for PSI are better than the result which is used in multi-regression. After the discussion with engineers for practicability of different model, we can find that clear value is not very meaningful for the engineering decision. After comparisons and discussions, fuzzy regression is more suitable when we cannot get more data in a short time. If we can collect generous data for a long-period, neural network is a better choice. The research applies the quadratic fuzzy regression to construct models. Because the quadratic fuzzy regression contains the attributes of necessity and possibility, it was identified with multi-regression. As a result, it has the same output with the traditional multi-regression and can be applied more flexibly.