Research on Bus Arrival Time Prediction Based on GTO-TCN-TFT Model

Accurate prediction of bus arrival times plays a crucial role in improving public transport operational efficiency, enhancing passenger travel experience, and advancing intelligent dispatching systems. Traditional time series forecasting methods often struggle to effectively capture the nonlinear spatiotemporal characteristics formed by the interplay of multiple factors such as passenger flow, road conditions, and traffic signal controls. To address these limitations, this paper proposes a hybrid GTO-TCN-TFT prediction model that integrates gated temporal units, temporal convolutional networks, and temporal feature transformers. This model is capable of fully exploring the deep temporal patterns embedded in large-scale bus trajectory data. Its multi-level cascade architecture effectively enhances feature extraction and long-term dependency modeling capabilities, while mitigating gradient degradation issues in deep network training through gating mechanisms and residual connections. To validate the model’s performance, the study utilizes actual bus operation data for experimental testing and analysis. Results demonstrate that the GTO-TCN-TFT model achieves higher accuracy and stability in bus arrival time prediction, showing significant advantages over traditional machine learning methods in modeling complex spatiotemporal nonlinear relationships.