Urban Road Network Traffic State Recognition Based on Road DNA Feature Sequence Matching

Urban traffic state recognition is of great significance for traffic management departments to timely grasp the status of traffic operations and formulate advance decision-making plans for urban traffic congestion diversion. To address the challenges of high identification accuracy and computational resource requirements brought about by complex traffic flows in urban road networks, this study proposes a traffic state recognition method based on road DNA feature matching. By describing road features in the form of DNA sequences, it achieves rapid matching of similar road sections and reduces repetitive training for similar feature road sections. Using detector data from Wuhan, we employed Bayesian optimization XGBoost hyperparameter model and trained with data from 31 road sections in the road network to construct a traffic state recognition library. Furthermore, we used the WH-KNN method to search for the trained BO-XGBoost models from the recognition library to identify the traffic state of the remaining road sections. Experimental results show that the recognition accuracy can reach 86.7%, and the training time of this model can be reduced by 61.2% compared to BO-XGBoost. Compared to traditional recognition models, both recognition accuracy and stability have been improved.