Research On Expressway Traffic Incident Detection Alogorithm Based On BOA-LSTM

As the most basic and extensive transportation infrastructure,roads have an irreplaceable and important function in the comprehensive transportation system.The "Outline of National Comprehensive Three-dimensional Transportation Network Planning" released on February24,2021 points out that in the new stage and new requirements,it is necessary to continue to promote the development of the national highway network in the direction of high quality.As an important part of the highway network,the total mileage of expressway has reached161000 kilometers by the end of 2020.Along with the rapid development of expressway,it also increases the probability of traffic accidents and makes the operation and management of expressway more difficult.Therefore,timely and accurate traffic incident detection can minimize the impact of traffic incidents and avoid secondary accidents,while also providing technical support to traffic management departments for safety management and improving road safety and service levels.Automatic traffic incident detection is an important part of Intelligent Transportation System(ITS),which is a classical 0-1 classification problem that can classify traffic states into two kinds: normal states and incident states.In real life,the occurrence of traffic incidents is episodic,the traffic incident state is much less than the normal traffic operation state,and the collected traffic data is unbalanced,so the traffic incident detection is essentially an unbalanced binary classification problem.Based on this,this paper will apply the imbalance classification technique to solve the traffic data set imbalance problem,and on this basis,a more comprehensive set of initial variables for traffic incident detection will be constructed,and then feature variables that are more sensitive to incident detection will be screened as the input variables of the detection algorithm,and the long short-term memory network(LSTM)in deep learning will be used as the classifier,and a hybrid sampling and feature variable selection based on the Bayesian optimized long and short-term memory network(BOA-LSTM)traffic incident detection algorithm based on mixed sampling and feature variable selection is proposed.Firstly,in order to solve the imbalance problem of the traffic dataset,a hybrid sampling technique based on the combination of Borderline-SMOTE oversampling and Tomek links undersampling is proposed,thus making the dataset balanced.Secondly,based on the theory of traffic flow fluctuation,the characteristics of the traffic flow changes on the upstream and downstream sections of the event occurrence point under the incident state are analyzed.Based on this,an initial variable set containing 15 variables is constructed by combining the basic traffic flow parameters in multiple perspectives and forms,and more sensitive feature variables are screened as inputs to the later model using the Random Forest-Recursive Feature Elimination with Cross Validation(RF-RFECV)algorithm.Then,the principle structure of long and short-term memory network is introduced,and it is applied to traffic incidents detection with hyperparameter optimization using Bayesian optimization algorithm to make LSTM have better classification effect and improve algorithm performance.Finally,the VISSIM simulation software is applied to simulate the scenario of a traffic incidents occurring on expressway to obtain the raw data of traffic flow and to train and test the performance of the algorithm proposed in this paper.The experimental results show that the traffic incident detection algorithm of BOA-LSTM based on hybrid sampling and feature variable selection proposed in this paper is has advantages in almost all evaluation indexes,the incident detection efficiency is higher,and the algorithm achieves a better comprehensive performance.