資料融合於異質性資料推估路段行駛速率之研究

Abstract

在ITS系統架構中，先進旅行者資訊系統(Advanced Traveler Information Systems, ATIS)為智慧型運輸系統的主要次系統之一，其功能在於藉由交通資訊的提供，發揮協助使用者有效決策、促進交通路網使用效率之功能。本研究主要目的在於發展合適的資料融合模式，用以整合各偵測技術蒐集的交通資料，分別考量不同來源蒐集之異質性資料特性，最終以最佳權重法進行結合，得到代表路段之融合速率，以提供使用者、交通控制管理單位較精確、可靠之資訊。 本研究針對偵測技術與其蒐集方式考量異質性資料各別特性，在資料分類處理過程做一調整，期求得一有效參考的平均速率值於交通資訊系統中應用。在模式驗證部分，本研究將透過路測現場調查的方式與理論融合速率數值進行驗證，而理論融合數值之產生係由異質性資料根據不確定性量測與最佳權重法之演算流程進行數值融合，再進一步評估理論融合數值對於實際車流行為描述之準確性。速率融合結果透過準確性之評估顯示，不論是長路段融合速率或短路段融合速率皆具有推估車流平均速率之能力，且此模式在市區路段的一般混合車流中具有較高的穩定度，因此本研究之融合速率模式可廣泛應用在長路段與短路段，採用實施內容則需視系統設計需求、目標數值型態以及提供用途等因素而定。 在實際應用中，若車輛偵測器回傳資料漏失、或計程車在單位時間內未經過該路段致使無法蒐集樣本、以及因通訊技術故障或臨時沒有回傳資料至資料庫等可能因素使得資料來源不齊全，以上原因可能導致系統出現資料來源漏失狀況，本研究將以同時段之歷史資料為替代數值，使得融合運算仍符合異質性資料處理過程，求得一提供使用者參考之平均速率值。

Real time traffic data acquisition has been the core and basis of all development of the advanced traffic management system (ATMS). For the goal of predicting traffic speed by spot and linear parameters, the traffic information sources in the framework of system should include roadside detectors data from taxi fleets as probe vehicles and historical data to generate traffic data for the main artery of urban area. According to the architecture, this paper describes the technique of using data fusion of active and passive information that combines adopt the spot and linear data to become their characteristics for estimation of traffic speed based on entropy and optimum weight that satisfies needs for all the potential public and private users. Through the development of traffic data fusion process, the proposed fusion model will be used for the data collection, fusion and analysis of traffic information. The process is composed of three consecutive computational steps. The first step is data screening that uses to reduce inaccuracy of outliers in system. The second step is to transfer data on the same basis – space mean speed, classify and individualize the data, and the latter can measure the probabilities in several sample sets and produce the entropy. The third step is to apply the optimum weight rule to generate weight allocation for travel speed from different sources. The experimental results reveal that good fusion results are obtained through the proposed data processing. This paper not only puts emphasis on reviewing the layout of surveillance technologies but finds the frequency and the applicative length of the main artery suitable model.