Research On Traffic Incident Detection Algorithm Of Expressway Based On Ensemble Learning

Traffic incident detection algorithm is the core of traffic incident detection.The traditional expressway traffic incident detection algorithm has the problems of low detection accuracy,long detection time and poor flexibility.It cannot meet the expressway traffic event detection in the background of massive,high-dimensional and multi-source traffic data.Ensemble learning is a hotspot of machine learning.It combines multiple base classifiers with better performance for learning and training,can efficiently and accurately process massive and multi-dimensional data,and effectively improve the performance of classification detection algorithms.Therefore,based on the fixed detector data and mobile detector data acquired on the spot of the expressway in Hangzhou,this paper develops a research on the detection algorithm of expressway traffic incidents based on ensemble learning.Firstly,based on the theory of traffic flow,the characteristics of the traffic flow of the road section under the occurrence of the event are analyzed,and the measured parameters,predicted parameters and combinations of the traffic flow obtained at the upstream and downstream of the road section are determined as input variables of the event detection algorithm.Aiming at the problem of excessive redundant feature variables in the combination of traffic parameters,an event detection variable selection algorithm based on RF and GBDT is proposed to select event detection feature variables.Secondly,study the ensemble learning theory,compare the application characteristics of each integrated learning algorithm in the classification detection problem,and propose a fast road event detection algorithm based on IPSO-XGboost single source data.Thirdly,based on the single-source data event detection algorithm,analyze the characteristics of multi-source traffic data,use GA to optimize the BP neural network to merge event detection data,and design a LightGBM-based multi-source data event detection algorithm for massive multi-source data Down City Expressway incident detection.Finally,we perform performance test experiments on the event detection algorithms of single-source data and multi-source data established in this paper,and compare the performance with common event detection algorithmssuch as RF,GBDT,Adboost,BPNN,SVM and KELM under the same conditions.analysis.The results show that the event detection algorithms for single-source data and multi-source data are superior to other event detection algorithms under the same conditions in terms of event false alarm rate,event detection rate,and average detection time of the algorithm.Among them,the multi-source data event detection algorithm based on LightGBM has the lowest false alarm rate,the highest accuracy rate and a short detection time,and the event detection effect is the best.