An Approach To Identify The Potential Hazard Locations And To Improve The Safety Situations For An Urban Arterial Network Being Designed

This paper combines the traffic safety theory with the road network design problem to analyze the techniques of identifying the potential hazardous locations and improving their safety situations for road networks being designed. The key technologies adopted in this study are neural network theory, cluster analysis, variance analysis and the network optimization technique. The context of the study includes the following four parts. They are: (1) the accident prediction models for a urban arterial system, (2) the methodologies to identify the hazardous locations and their corresponding probable causes for a road network being designed, (3) the approaches to establish the safety level of service and to evaluate the traffic safety situations for a urban arterial system, (4) the ways to improve and to optimize the network being designed based on the consideration of traffic safety elements.First of all, the data needed for the study are analyzed and summarized. Then based on a carefully arrangement for data collection, a large quantities of traffic surveys and other field investigations are performed. During the process of data collections, the technique to determine the OD matrix is introduced and improved to retrieval those traffic volumes that can not be surveyed directly. This study tries to store and retrieval data by the software of TransCAD widely used in transportation planning field, and the results are agreeable.With respect to the study on accident prediction models, the mission to establish the accident prediction models is defined and appropriate structures of the models are discussed firstly according to the goal of the whole research work. Then, the links and nodes (i.e., the intersections) of urban arterial system are classified by applying the cluster analysis method, and the characteristics of the statistical distribution of the accident data are obtained. Furthermore, a set of muliti- non-linear regression models, the Logistic curves, are estimated to predict the frequencies of accidents occurred on each kind of links and on nodes, respectively. Lastly, the relationship between the accident indexes and the v/c during the traffic peak hour period is analyzed. The accident prediction models and the relationship models between accident and v/c can provide accurate and reliable accident data for the safety analysis on a road network which is being in its planning planning stage.Secondly, the merits, limitations as well as the suitable applications of each methodology to identify the hazardous location are summarized and analyzed. Then, the essence to identify the hazardous locations is revealed with the help of a case study. Based on these fundamental research works, the BP neural network models to identify the potential hazardous links and the potential hazardous intersections of the road network being designed are established and calibrated, which are characterized by the following three advantages when compared with the traditional ones: firstly, they can identify the potential hazardous sites of a road network before the construction of it; secondly, the factors or the input variables taken into account are all the variables that can be obtained from the outcomes of the road network design; lastly, the number of hazardous sites need not to be restrict to some limited amounts and it is determined by the rule that no risk sites are missed to be checked. With respect to hazardous sites identified, an approach to distinguish the overwhelming factors that cause the traffic accidents is presented based on the probability theory.Furthermore, the methodology to determine the safety evaluation criteria for links and nodes of the road network is put into forward according to the classification theory, and then the corresponding criteria are established based on accommodate traffic volume, accident frequencies and the equivalent injuries caused by accidents. With the establishment of the safety criteria for links and nodes, a safety index of the whole network is defined based on the criteria given above and the corresponding safety evaluation criteria for the network are presented according to the magnitude of the safety index.Thirdly, the structure and the mathematic description of the road network are analyzed when the traffic safety factors are taken into account. Then, the methods to evaluate the traffic safety situations of the road networks being planned and then to make some corresponding improvement of them are suggested, which have the functions of predicting the accidents, identifying the potential hazardous locations and their accident causes, evaluating the traffic situations of each links and nodes, and assessing the traffic safety situation of the whole network. With respect to the quantitative network design problem, a bilevel programming model which can minimize both the travel time and accident frequency is proposed for urban road network. Then, the way to convert the accident into equiverlent travel time, and the algorithms, which can be used to solve the model , are all discussed. Finally, an attempt is performed to take the predicting accident frequencies occurred on peak hours as part of impedance and to assign the traffic volume to the network according to the integrate impedance of travel time and accident frequency.At last, a practical network obtained from Harbin urban arterial system designed for the target year of 2010 is taken as a study case to illustrate the use of the set of methdologies established in this paper. The application of the traffic assignment on the same network based on the integrate impedance of travel time and accident frequecy during the peak hour is examined at the same time.The results of the study focus on the following six aspects showed below:(1) A accident prediction model bank composed of 24 models is established to provide the traffic accident data for the safety study on the road network which is being in its planning stage.(2) The methodologies to identify the hazardous links as well as the hazardous intersections are presented based on BP artifical neural network which can be used to identify the potential hazardous sites of a road network being designed.(3) An approach to identify the overwhelming factors that cause the traffic accidents occurred on the potential hazardous locations is given based on the probability theory.(4) The traffic safety criteria for each kind of link and intersection are established firstly. Then, a safety index of the whole network is defined based on the criteria given above. And lastly, the corresponding safety evaluation criteria for the network are presented according to the magnitude of the safety index.(5) The methods to evaluate the traffic safety situations of one or more road networks being planned and then to make some corresponding improvent of them are suggested, which have the functions of predicting the accidents, identifying the potential hazardous locations and their accident causes, evaluating the traffic situations of each links and nodes, and assessing the traffic safety situation of the whole network.(6) A bilevel programming model which can minimize both the travel time and accident frequency is proposed for urban road network, and the algorithms which can be used to solve the model are discussed.