應用分類元方法於微觀車流系統之模擬－以高速公路多車道為例

Abstract

基於理論基礎所建立車流模型與交通動態預測系統，可作為交通管理者的決策分析工具，並提供即時的交通路況資訊之服務。車流於ITS中之應用除了要求模式之正確性以求能符合現實情況外，另一關鍵在於其反應速度。現實中的車流狀況隨時間而有所變化，且其情況通常相當複雜，以致其車流模型在運作上常需要相當大的處理時間及運算能力，故需要在運算時間及正確性中取捨。近年來，演化式計算（Evolutionary Computation）逐漸成為人工智慧的一個重要研究領域，且被廣泛運用於需要複雜運算的各領域研究中。演化式計算具有許多優點，例如具有機器學習的特色、可解決許多領域問題之彈性與適應性，並具有全域解搜尋的能力，因此非常適合作為解決困難的最佳化問題之工具。演化式計算的具體實現，除了廣為人知的基因演算法與基因規劃之外，還有同樣主要是由Holland 所發展的分類元系統（classifier system）。 交通控制與管理應先行了解車流行為特性，若能有效掌握車流型態，即可藉由先進科技的應用來提升運輸系統效能。利用先進科技蒐集取得真實交通資料，透過分析以瞭解與掌握車流行為特性，進而加以模組化，將可作為交通管理者的決策分析工具，亦可提供即時的交通路況資訊，為實現智慧型運輸系統所不可或缺的重要環節。本研究嘗試應用分類元方法於微觀車流模型，冀望能應用分類元所具有之高適用性及快速反應性在車流模型上，於可接受的時間內得到一具實用性的高快速公路微觀車流模型。本研究於初步研析高速公路車流特性、道路幾何特性，車流特性以及駕駛行為，並進行分類元法相關文獻回顧以了解其理論基礎、特性及應用方法後，進一步研擬如何應用分類元方法於微觀動態車流之預測上，將所得結果以分類元法應用於國內高速公路小客車車流模式之建立，並藉由實例驗證，以校估及驗證所建立模式之準確性，冀望能發展一可於有效時間內得到結果，且能符合實際車流情況之模式。

The traffic flow model and dynamical. predict traffic system base on theoretical foundation would be the traffic administrator’s decision tools and could offer the service of the traffic road conditions information immediately. The traffic flow becomes the application in ITS besides requiring the exactness with the reality, another key lies in its reaction speed. The flow state in reality changes with time, and its situation is usually quite complicated, so that traffic flow model often needs long time in operation. In recent years, Evolutionary Computation is an important research field in artificial intelligence, and is applied to solve problem which needing complicated operation. It has a lot of advantages, for example, like machine learning which can solve a lot of elasticity and adaptability problem. Beside the gene algorithm is one of widely known algorithms was developed by Holland, the classifier system has same to be. In order to control and manage traffic, it should understand car flow characteristic in advance. Using advanced science and technology to collect and make the true traffic data, and through analyzing the characteristic to understand and get the car flow’s behavior. Then it can be the analyzing tools for administrator's decision of the traffic, and can also offer the traffic road information immediately. This is the important link with indispensable in intelligent transportation system. This study tries to apply classifier system on microscopic traffic flow model, and hope to get a practicability high way microscopic traffic flow model during acceptable time. It will do a research about the characteristic of car flow, the characteristic of road geometry and the drivers’ behavior in highway in advance, and then do paper review about classifier system in order to understand the theoretical foundation, characteristic, and applying method. And more, try to find how to apply classifier system to predict microscopic traffic flow, and then construct the Microscopic traffic flow with classifier system method. According to the case verify result, it could be used to correct and modify the model to get more corresponding output result. At last, we hope to be able to develop the model which can receive result in effective time, and be correspond to reality traffic flow of situation.