Urban Traffic Level of Service Prediction Method Based on DBO-Transformer Multi-Source Information Fusion

This study proposes an advanced framework for scientifically predicting the traffic level of service (LOS) that overcomes the constraints of conventional methodologies dependent primarily on real-time vehicle speed and traffic flow metrics. Recognising the empirical challenges in transformer model parameter optimisation, which is typically set based on engineering experience, we develop a novel dung beetle optimiser (DBO)-optimised transformer architecture. The model utilises traffic flow data from Fuzhou urban expressways to systematically optimise two pivotal hyperparameters: the number of self-attention heads and the learning rate. The model’s performance is evaluated using mean squared error as the primary metric, supplemented by experimental data visualisation for a comprehensive assessment. The study investigates the model’s predictive accuracy for the next moment’s level of service under various influencing factors, including vehicle speed, traffic flow, time, weather, holidays, temperature and traffic conditions. The results indicate that the DBO-transformer model, which integrates multi-source information fusion, achieves exceptional performance in traffic level of service prediction, with an accuracy of 98.0495%. This performance surpasses that of the standard transformer model, the DBO-optimised BP neural network model and the LSTM neural network model.