熵應用於交通資料融合之研究

Abstract

在許多智慧型運輸系統(ITS)的應用中，即時資訊的需求越來越高，為了能提供正確可靠的資訊，ITS應用所提供的資訊必須持續更新，意謂著資料收集與處理過程必須持續進行。然而，由於資料來源(如偵測器、探針車等)有偵測範圍的限制且資料量通常不大，易造成資料的可靠度降低。藉著資料融合的過程，可以改善以上問題。 資料融合的觀念始於1980年左右，而在近年才有許多資料融合方法的發展與應用。回顧在ITS領域中相關的資料融合研究，主要可以將這些技術分成三個層級，其中，層級二的資料融合技術能夠提供由原始資料而來的推論以及更完備資訊，因此本研究的目的是在發展層級二的資料融合模式。 熵值的應用是由C. Shannon在1948年所提出，最早被用於“信息理論”，之後，熵值被廣泛地應用於量測不確定性，本模式中，提出了資料分級的方式使得熵值可以應用在量測各資料來源的不確定性。而由於熵值所代表的含意為不確定性，模式也更進一步推導權重與熵值的關係，給予每個資料來源最佳的權重。 為了測試模式的適用性，我們設計了一連串的實際測試。而在資料收集不易以及資料量不大的情況下，我們亦使用模擬的資料來進行測試。測試結果證實本研究所提出的資料融合方法在實務上具有可行性。

Real-time travel information is becoming increasingly important in many intelligent transportation system (ITS) applications. In order to provide reliable information to the users, traffic information in all the ITS applications should be comprehensive and continually updated. It means that a continuous real-time data collection and processing effort is essential to provide the required information. However, data sometimes is not reliable since every source has a certain detecting range and the data volume is often small. These problems can be addressed by data fusion process. Data fusion technology started in the late 1980s and many data fusion approaches had been developed and applied in recent years. In reviewing data fusion techniques in ITS field, the techniques can be divided into three levels. In our model, we propose data fusion techniques focus on the level two since level two data fusion provides a higher level of inference and delivers additional interpretive meaning suggested from the raw data. Entropy is a concept proposed by C. Shannon in the 1948 and is used in “ Information Theory” first. Shannon’s entropy function has been used extensively as a measure of uncertainty. We propose a classifying approach so that we can cite the entropy to measure the uncertainty of the collected traffic data. Since entropy represents the uncertainty, we form an optimal weight scheme and use entropy to derive the weight of each sensor. We perform a series of tests for model evaluation purpose. Since collecting real data is hard in practice and the volume of real data is often small, we also use simulated data to test our model. The testing results show that our proposed entropy data fusion technique is suitable in practice.