Development And Evaluation Of W-CUSUM Incident Detection Algorithm Based-on Mobile Source Data For Urban Expressways

Traffic incidents are a primary contributor to various unexpected congestions, resulting in traffic delays, declining of traffic infrastructure service efficiency, and loss of traffic users' benefits, and having a considerable impact to the traffic safety and environment as well. A prompt and reliable detection of and a quick response to the traffic incident on urban expressways, which represent the trend of the rapid urban infrastrure development and the backbone of the urban transportation network, is not only an important component of traffic management and control of urban expressways, but an effective way to decrease the losses resulted from the incident and reduce the traffic congestion as well. Meanwhile, the continuing development and maturity of mobile traffic detection techniques are now able to provide new real-time, dynamic and accurate traffic information for incident detection.The purpose of this dissertation is to develop, evaluate, validate and implement an automatic incident detection (AID) algorithm for urban expressways based on mobile source traffic data, which provides a new approach for the detection and management of, and response to the traffic incidents on urban expressways as well as an evaluation method of AID algorithms. The research in this dissertation consists of the following accomplishments and contributions:1. Existing AID algorithms are mainly developed based on fixed detectors, which have such shortcomings as high false alarm rate, and poor application performance. In this context, this research proposes an AID algorithm, called W-CUSUM algorithm, for urban expressways based on mobile source data. In the proposed algorithm, the feature of the mobile source data and the temporal-spatial characteristics of the traffic flow on urban expressways under incidents are fully considered by combining the wavelet analysis method and CUSUM approach, which improve the quality of the data, thus greatly increase the accuracy and reliability of the incident detection.2. The existing false alarm rate index could hardly reflect the false alarm situation directly and effectively, and the traditional performance envelop curve can only give a qualitative and rough description of the performance, which result in a lack of comprehensive and quantitative evaluation of the AID algorithms. Therefore, this research first proposes a new false alarm rate based on temporal and spatial index, and then establishes a comprehensive index based on expected cost by integrating the indices of detection rate and false alarm rate. The proposed index based on expected cost can not only provide the comprehensive and quantitative evaluation and comparison for various AID algorithms, but be used for determinating the optimal detection threshold and performance point of an algorithm as well. Thus, it gives traffic managers an effective technical approach to compare, select and calibrate the AID algorithms.3. The proposed W-CUSUM algorithm is evaluated and compared by using both simulation data and real-world data. By analyzing the performance of W-CUSUM algorithm under different influential factors and comparing W-CUSUM algorithm with SND, CUSUM and UCB algorithms, it is found that:(1) the number of blocked lanes has the largest effect on the performance of W-CUSUM algorithm, while the incident duration has little effect on the performance of W-CUSUM algorithm; (2) W-CUSUM algorithm performs better than SND, CUSUM, and UCB algorithms with an improvement of 33.7%,19.5% and 38.3% respectively.4. A model for determining the mobile source data sample size for incident detection is developed based on the Experimental Traffic Engineering Methodology (ETEM), which overcomes the shortcomings of existing methods. According to the simulation case study, the reasonable percentage of floating cars in road networks and the optimal detection interval under certain accuracy of incident detection are obtained.5. A multi-source data fusion model for incident detection based on hierarchical structure is developed by fully considering the characteristics of the incident information contained in various available data sources and integrating W-CUSUM algorithm, neural network model and the experiences of traffic decision-makers. The proposed model not only enhances the flexibility and expandability of W-CUSUM algorithm for application, but also provides a pratical methodology for improving the accuracy and reliability of the W-CUSUM algorithm.6. A framework for implementing the W-CUSUM algorithm is established using a systematic and integrated approach, in which the development, installation, and operation of the AID system are presented in details in order to provide an overall methodological system for the W-CUSUM algorithm to transition from a theoretical study to practical applications.