Modeling And Simulation Of Freeway Incident Detection Based On Information Fusion

Freeways play an important role on providing a safe, efficient, comfortable and convenient fast way, also great social and economic profits. With the dramatically increasing on surface transportation networks, numbers of cars, congestion and all kinds of incidents on China highways continues to be a serious problem that is growing steadily worse. Incidents are major contributors to congestion, personal injury, property damage and costing in lost productivity.Efficient and reliable incident detection technologies and operational procedures will reduce the time a freeway incident affects traffic and will lead to substantial reductions in motorist delay and damage. The key is to detect incident as earlier as possible, identifying incident characters with timely succor and providing interrelated information for drivers, which is needed to be established a perfect incident management system, managing incident fast, efficiently and properly. Incident detection is considered to be the fist step and the most important step of incident management system. Incident management system consists of sensor data collection system and detection algorithms. Incident detection algorithm is the system key part except for sensor data collection technologies.Many Automatic Incident Detection (AID) algorithms are developed in the past 40 years. AID is considered to be a potentially reliable technology that many AID algorithms have been applied widely by both academics and practitioners in developed countries. Given the variety and dimensional complexity of these algorithms, single algorithm is very hard to achieve the best detect effects. Algorithm fusion is able to improve the performance of detection. It is necessary to do some research on algorithm fusion of incident detection problems and to design effective algorithms for it. Recently, traffic sensor collection technology grow quickly with the development of electron, sensor, communication and computer technology. It is necessary to do some research on incident information fusion of incident detection problems and to overcome the shortcomings of single incident information. Although, it is late to research on incident management system and AID algorithms for domestic researchers, some fruits have been achieved. Until now, few researches have been made on algorithm fusion and information fusion of incident detection problems, and many dissatisfactory items await amelioration and modification. In this dissertation, a series of modeling and simulation of freeway incident detection based on information fusion are analyzed thoroughly.Accelerating the incident detection research of incident management system will speed domestic incident management level on the one hand and enrich application study of information fusion on traffic engineering.Firstly, leading in information fusion technology, discuss the input and output characteristics of information fusion process. We construct a generalized framework for information fusion incident detection with five basic kinds of fusion process and reclassify AID algorithm based on the framework. Based on the concepts of entropy, conditional entropy, average conditional entropy and mutual information, we prove feasibility and validity of information fusion used in incident detection from the angle of information theory. Based on this, we discuss modeling and simulation of the ANN and SVM incident detection models, voting fusion of incident detection algorithm and information fusion D-S method of incident detection technology in detail. The main contents are as follows:1. Based on traffic theory for incident detection, we use failure diagnose method to discuss fundamentals of Artificial Neural Network (ANN) and Support Vector Machine (SVM) which are a FEI-DEO fusion process in traffic incident detection.2. A Multi-Layer Feed-forward neural networks (MLF) model, a Probabilistic Neural Network (PNN) model and three SVM models with different linear, polynomial and radial basis function for incident detection are presented.3. Based on comparison of the three kinds of incident detection models from academic point of view, we validate and compare the three models by worked loop data and incident data of 1-880 real field database. A univocal result is achieved about field out-line incident detection modeling of MLF, PNN and SVM on 1-880 freeway segment.4. In order to solve incident detection algorithm fusion problem, we introduce a voting fusion method to incident detection and propose a voting fusion method of incident detection algorithm fusion. Essential concepts of voting fusion method are analyzed and modeling steps are established.5. A Monte Carlo simulation model for incident detection results is found. Based on simulation, incident detection simulated results are produced and provide data for case analysis.6. A case from MLF, PNN and SVM three models shows that voting fusion method of incident detection algorithm fusion is effective. Four key problems: combinations, confidence levels, detection modes of algorithm voting fusion system and incident detection table, Boolean algebra of detection and false alarm probability are solved.7. A method of hardware implementation for voting fusion method of incident detection algorithm fusion is proposed.8. Based on combination of incident technologies, in order to resolve incident detection technology information fusion problems, an information fusion D-S method of incident detection technology is established. Three key problems: information sources, value of basic belief function and decision method of information fusion D-S method of incident detection technology are decided.9. Analyzing validity of information fusion D-S method of incident detection technology based on proving five propositions.10. A case from Inductive Loop Detectors, Cellular Phone and Close Circuit TVs shows that D-S method of incident detection technology information fusion is effective.