A novel deep learning method for intelligent quantitative evaluation of urban road ride quality under complex road conditions

Driving comfort is a key indicator of road quality. To address the limitation of ISO standards in accurately quantifying ride comfort under complex road conditions, this paper proposes a ride quality evaluation method that accounts for urban road complexity. First, a self-developed sensing device is used to collect vehicle vibration acceleration and speed data. Second, an adaptive deep learning model with a multi-head attention mechanism is developed, using the Maximum Transient Vibration Value (MTVV) for extreme road segments and the Weighted Root Mean Square Acceleration (WRMSA) for normal segments as evaluation metrics. Finally, road quality is characterised by the expected value over spatial segments. Results show that the proposed vibration data collection method is reliable, and the model achieves an R2 of 0.92 compared with ISO-based calculations. This study provides support for refined maintenance and management of urban roads.