A Knowledge Based Real-Time Travel Time Prediction System using Location Based Service

• Main Authors: Sheng-Han Tsai, 蔡昇翰
• Other Authors: Shian-Shyong Tseng
• Format: Others
• Language: en_US
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/28880313129401455539

碩士 === 國立交通大學 === 資訊科學與工程研究所 === 94 === The purpose and the essence of developing Intelligent Transportation System (ITS) are to utilize advanced communication techniques, traffic control and information to achieve a convenient, economic benefits and safety traffic environment. In ITS area, real-time travel time prediction (TTP) topic has been discussed recently, because this important topic covers four of nine research subjects in ITS domain. Such as：Advance Traffic Management System, Advance Traveler Information System, Commercial Vehicle Operation and Emergency Medical Services. Also, it presents an index of real-time traffic condition and useful traffic information. However, most previous researches focus on the predicting the travel time on freeway or simple arterial network. The real-time TTP in urban network is hard to be achieved in four reasons: complexity and routing problem in road network, sensor data is either not available in real time or is not cost-effective to get in real time, spatiotemporal data coverage problem of sensor based or vehicle based travel time prediction, and lost precision because lack of traffic event response mechanism. In this thesis, the knowledge based real-time TTP system is proposed, which uses data mining technique to discover some target traffic patterns/rules with location based service (LBS), and then uses inference engine with previous traffic pattern/rules and real-time traffic information to predict the real-time travel time. When traffic events occur in some road sections, the meta-rules are triggered by the system to dynamically combine real-time and historical travel time predictors. The proposed system is implemented for Taipei urban network, and experiment results show that weighted combination of real-time and historical predictors outperforms either single predictor.