Research On The Prediction Of The Severity Of Accidents On Mountain Expressways Based On Particle Swarm Optimization BP Network

Most freeway accidents are traffic anomalies that occur due to multiple factors such as people,vehicles,roads,and the environment.Large number of researches point out that people are the main cause of freeway traffic accidents,but it is undeniable that there are still exist traffic accidents that are directly or indirectly caused by other factors such as complex road conditions or poor weather conditions.Such situations are very common in mountainous regions of Yunnan,the complex linear conditions and the changing weather conditions in mountainous regions can easily lead to serious traffic accidents.Therefore,it is of great practical value to study the traffic accident prediction model for freeways in mountainous regions.In view of this,it is expected that a certain rule can be extracted from the traffic accident records of a great many of existing mountainous regions freeway to predict traffic accidents on freeways in mountainous regions,and to better identify dangerous road sections on mountain freeways,give drivers better instructions to avoid accidents as well as improve the safety of mountainous regions freeway.The relationship between line shape and number of accidents can explain the occurrence rule of accidents to a certain extent,however,the occurrence of traffic accidents is often caused by multiple influencing factors,studying the relationship between a single line and the accident rate does not accurately explain the occurrence of the accident.Based on this,environmental factors such as time,weather and road surface conditions in the accident records and vehicle factors such as vehicle types are put into input variables,and model for freeway accident severity grade prediction in mountainous regions is established.This paper selects K0 to K78 section of Kunshi Expressway for the research,and more detailed collection and collation of traffic accidents,road alignments and related data have been done.The historical accident data in this paper are from historical traffic accident data of the Kunshi Expressway from 2008 to 2012.According to the network training results of BP neural network,the relationship between influencing factors and traffic accidents is about to obtained.The BP neural network training algorithm is optimized.The particle swarm optimization algorithm is selected as the network learning algorithm.The MATLAB neural network toolbox is used to implement the prediction model at last.This paper briefly introduces the principles of BP neural network,network structure and the principle of particle swarm optimization algorithm and basic process of algorithm.It also states the process of implementation of BP neural network optimized by particle swarm algorithm and its network and its implementation method of network modelFirstly,collating the raw traffic accident data records on Kunshi Expressway,and preprocess the raw data.The processing of raw data mainly includes the selection of sample data as well as the quantification of each influencing factor of the sample.Then,the classification of the input layer and the output layer is carried out on the preprocessed accident data.The factors that affect the accident are classified as the input layer,the factors that can represent the severity of the accident are classified as the output layer,and the “influencing factor-accident severity level” format is obtained.The input layer includes: the time of the accident,the weather condition at the time of the accident,the type of vehicle,the condition of the road surface,and the flat curve radius and vertical slope of the place where the accident occurred;the output layer is the number of casualties in the accident.Using neural network toolbox programming of MATLAB to realize the establishment of network prediction model.Secondly,put the pre-processed 312 accident sample data into the network model for training,and the relationship between the influencing factors and the severity of the accident is analyzed through the prediction results of the model.Finally,sample the remaining 52 groups of real accident data from the sample data and import them into the prediction model for prediction and verification.The model’s predicted data and real accident data are analyzed and compared,the accuracy rate reached 89.6%,indicating that the model can be used to predict the severity of accidents.Finally,on the basis of the prediction model,a set of situation simulation simulation is performed to predict the severity level of the accident in this scenario.The law of accident occurrence is obtained from the prediction results,and it is found that serious accidents are more likely to occur under the following conditions.: Driving time is 0:00-8:00 in the early morning;Weather conditions are rainy and foggy days;The radius of the road curve is 0-600 meters;The type of vehicle is heavy truck,etc.