A deep learning method for assessment of ecological potential in traffic environments

Abstract

To further enhance the energy efficiency of the road traffic system, this study comprehensively considered various factors such as road conditions, traffic situations, and weather environments, extracting a total of 34 feature variables affecting the ecological nature of traffic scenarios. A feature selection method combining Random Forest, Permutation Importance, and Sequential Backward Selection algorithms was used to determine the optimal set of features, which includes 12 variables. Subsequently, a traffic scenario ecological characteristic assessment model based on the Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) algorithm was constructed to improve the overall performance of the road transport system. By testing and comparing eight deep learning algorithms, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and CNN-LSTM, the effectiveness of the constructed model was verified. The results indicate that the CNN-LSTM algorithm performs best in the ecological assessment of traffic scenarios, capable of accurately classifying all features, with Accuracy, Precision, Recall, F1-score, Micro AUC score, and Macro AUC score reaching 0.83, 0.826, 0.83, 0.825, 0.945, and 0.904, respectively. Additionally, this study employed the SHapley Additive exPlanations (SHAP) method for interpretability analysis of the model and used violin plots to demonstrate the distribution of various features across different scenario categories. The results show that the type of functional zoning to which the road geographical location belongs, visibility, and various traffic condition features have significant correlations with the ecological category of road traffic scenarios. Therefore, appropriate traffic energy-saving and emission reduction control strategies can be adopted for different functional zones, weather conditions, and traffic situations to promote the road traffic sector towards a zero-carbon goal.