Pattern Recognition Of Traffic State At Signalized Intersections

With the continuous increase in the number of motor vehicles,the problem of traffic congestion has become increasingly prominent,which has seriously affected the operational efficiency of the transportation system.Traffic state recognition is a prerequisite for traffic control and guidance,which can effectively alleviate traffic congestion,so it is of great significance to study it.Based on the traffic characteristics of signalized intersections,this paper establishes the traffic pattern recognition model of intersections,and formulates the signal timing scheme based on the recognition results.In this paper,by drawing the time sequence diagram of the traffic flow at the intersection,it is found that it has volatility and periodicity,and based on this rule,a traffic state recognition model for the intersection is established.The recognition model believes that the combination of different traffic states in each phase constitutes different traffic patterns at intersections,and the traffic state in each direction of the intersection is the basis for recognition.At the same time,combined with the traffic state measurement methods of road sections and intersections,this paper divides the traffic state of intersections into four categories: the smooth,the stable,the congested and the seriously congested.Combining the commonly used identification parameters and the traffic characteristics of the intersection,we select the queue length,saturation and occupancy rate are used as traffic parameters for traffic state pattern recognition,which lays the theoretical foundation for clustering and recognition below.The identification of the traffic state in each direction of the intersection adopts the method of combining unsupervised learning and supervised learning.According to the fuzziness and uncertainty of the traffic state,this paper applies fuzzy theory to the clustering algorithm.Considering that the clustering result of the fuzzy C-means is more dependent on the selection of the initial clustering center,genetic algorithm is introduced,and the serial number of the traffic data sample is used as the coding object,and the initial clustering center is optimized to obtain a more stable and accurate traffic state data set.The method is verified by traffic data simulated by VISSIM,and the result shows that the GA-FCM traffic state clustering algorithm performs better than the FCM algorithm in the global search ability,and the number of iterations is reduced by 38.71%.In order to complete the real-time discrimination of traffic state,this paper proposes a limit learning machine model based on particle swarm optimization.The initial weight and deviation are determined by particle swarm optimization algorithm,and then the machine learning network is trained by using the historical data corresponding to the four traffic states.The result shows that the PSO-ELM algorithm has better recognition accuracy and stability than traditional ELM,KNN and SVM algorithms.Finally,the expert scoring method is used to describe the degree of demand for green light time in the four traffic states,and the weights of different traffic states are obtained.Based on the weights,the timing schemes for different traffic patterns are re-formulated.Then,taking the average queuing length,the maximum queuing length and the average delay time of vehicles as evaluation indicators,the VISSIM simulation experiment shows the feasibility of the signal timing scheme considering the results of traffic pattern recognition.