Vehicle Behavior Model And Accident Detection Influenced By Expressway Accedent

Expressways have been playing an important role in promoting economic construction and social development of China in recent years. Whereas efficient, secure, unimpeded transportation should be guaranteed by ATMS (advanced transportation management system). Even as stated by ITS (Intelligent Transportation System) User Service System, EAM (expressway accident management) is essential for establishing ATMS, the key technology of which is accident detection.In view of existing national transportation situations, Vehicle Behavior Model and Accident Detection Affected by Expressway Accidents is a research exploring on the key technology systematically. Based on reading abundant documents and extensive investigation, the study is performed according to the 3-step method: vehicle behavior model analysis and reconstruction, traffic simulation, and incident detection.1. Through investigating vehicle behavior in expressway accidents and analyzing the influence on expressway traffic current caused by accidents as well as various behaviors under the influence, occurring circumstances are identified for 4 types of vehicle behaviors. Car-following Model Numerical Simulation System has been developed and leading car behaviors have been defined under different effect degree caused by accidents. On account of the assumptions, the existing models have been analyzed as numerical modeling, reflecting the car-following process. Characteristics of each car-following model and existing problem have been represented, which provide scientific basis for model reconstruction. Furthermore, car-following experiment has been carried out by utilizing Five-Wheel Meters, and datum of car-following has been obtained. Lane-changing behaviors influenced by accidents have been studied, and the conditions for forced lane-changing behavior and voluntary lane-changing behavior have been drawn, and the four types of terms for lane-changing behavior as well as the existing model have been studied, which provide experimental, theoretical and practical basis for reconstructing model.2. Leading in Agent and MAS theory, the paper compares agent and vehicle behavior and also combines each other, which lays the foundations for introducing agent and MAS theory. Vehicle-Agent Structure Model and car-following model frame have been established with the car-following model based on agent being suggested. The BDI (Belief→Desire→ Intention) model has been projected; meanwhile each array and element has been analyzed. Particularly the Headspace-desire in D_F-model has been discussed quantitatively thus headspace-desire model has been set up. By means of PD control, the acceleration effects caused by relative rate and space respectively have been defined. Moreover, I_F-model has been described in detail.In addition, in accordance with Vehicle-Agent Structure Model, lane-changing model based on Agent has been set up. BDI property analysis of the model has been provided, with the B_C model, D_C model and I_C model being established. Especially four rules for lane-change have been obtained:draw space relationship of lane-changing and form interest function which gives quantitative description of space relationship; abstract time rules and set up time-rule model concerning lane-changing; form the state rules of lane-change of Vehicle-Agent and determine the desire rules of lane-change through quantitative analysis which depicts the subjective ideas quantitatively.Based on what have been mentioned above, the research has verified, evaluated and proved the fact that the model established in research under the influence of expressway accident is of high performance.3. In order to seize the traffic simulation characteristics influenced by expressway accidents, entity attribute is analyzed to VehicleAgent, and then ER model is established. The demand analysis idea based on laminated-analysis is brought forward. The demand analysis model based on laminated-analysis of the problem is established. Considering the problem, the cases of demand analysis of laminated analysis are given. The relationship among Vehicle_Agent is built, and the communication of MAS system is described. The communication contents, type and negotiating agreement are analyzed, and the principles for compromise settlement are pointed out. At the same time, the relationships among different controlling areas are analyzed, which lays the logical foundation for making further simulation. To ensure the veracity and accuracy of simulation, part of the expressway driver attributes, traffic composition, traffic flow parameter and expected speed of our province and even Shanxi and Yunnan are investigated. Problems about traffic production and randomness are discussed and AID—Simulation system is developed. Based on that, the combined data demarcate is studied, and the expressway traffic database of accidents is set up, which insures the variety and typical of data4. The original evaluating indicator is improved, ReR (Recognition Rate) target is added, and EDR (Error Detection Rate) index is raised. The changing features of expressway traffic flow influenced by accidents are studied and the properties of problems are fixed, which foreshadows the introduction of MLF for event recognition. The accident test model of MLF is installed and the basic structure and test type of model are set. The characters of test parameter unilateral input and two lateral input of test parameter are analyzed quantitatively and then a scientific conclusion is drawn. Moreover, the combination of single factor and multifactor on input characteristics is studied and the relationship among various factors and test efficiency is gained. The characteristics plan for different combination input is set up and function test for different combination is obtained. And on the basis of that, input neural units' characteristic plans of different time combination and different factor combination are fixed. The model's hide layer structure is studied, and the optimal unit structure is gained. To insure the correctness of selected models, SOFM and ART2 are introduced contrastively and accident test model is established by means of this theory. The comparative analysis shows that the test model based on MLF expressway accidents has the best comprehensive test function.5. In the light of the extensive applying prospect of SVM pattern recognition, linear SVM has been introduced and classified experiments have been conducted according to specific problem; in a similar way non-linear SVM and high dimensional basic theory have been led in and pattern recognition experiments have been carried out. Theperformances of different cores have been discussed and enforcement of pattern recognition has been studied. By studying the changeable features of traffic current parameters under the influence of expressway accidents, separability, linearity and non-linearity have been explored and the basic scheme for classifying has been fixedThe expressway accidents detection model based on SVM has been set up and detection has been accomplished concerning specific problem. According to those experiments, the basic feature has been analyzed. Aiming at one-way and both-way in-put, the detection performance has been studied; as for combination of different input performance factors, experimental analysis has been conducted with performance index for different detection programmer. With the influence brought by different cores on detection performance, the core agreeing with the problem is determined.6. Introducing wavelet analysis theory, expressway accident detection model based on one-dimensional has been established. For the purpose of comparative analysis, expressway accident detection models which make use of one-dimensional discrete transformation and two-dimension transformation are set up. A conclusion for applying one-dimensional discrete transform to detect incidents is reached. Expressway accidents detection model based on one-dimensional discrete transformation is built up with the fundamental procedure of detection; characteristic variables of patter recognition model are formed. The paper has put forward basic recognition patter and has examined the recognition patterns. After making comparative analysis of detection performance of different wavelet system and different wavelet in the same wavelet system, a proper wavelet basis is found and the scale change is analyzed simultaneously.7. Considering diverse conditions in accidents, such as lane, detection space, sampling period and database volume and so on, the study sets up database containing 8 different happenings. With the help of the database, the three types of model and California 8# algorithm are all analyzed comparatively. The conclusion that the models in this research are of high performance has been drawn. The comparative research of the three models has been done. In order to improve comprehensive performance of detection, MSW secondary combination detection model has been established according to combination detection of 3 types of models. Comparison of detection efficiency between the model and the other four existing models proves that the model in this study is of high detection efficiencyBesides, the study on online detection time of accident has been carried out, through which the fundamental law that average detection time varies with space change and information sampling periodically has been acquired; Finally the paper compares the accidents on single and dual lane.The research was sponsored by Sichuan Traffic Technology Key Project. (2003-7-10).