Congestion Prediction Model for Toll Plazas Using GRU and Optuna Approach

The escalating incoming strength of vehicles can result in extensive queues or congestion at toll plazas. This congestion at the toll collection lanes has a significant impact on the flow of traffic, which can severely impede the efficiency of the expressway toll plazas. One of the crucial elements in toll plaza management is the capability to predict congestion at the toll plazas. Current studies are encountering limitations, such as inaccurate data and failure to reflect actual traffic conditions. This study proposes a new model of congestion prediction for toll plaza management. The model adopts four calculated parameters including average queue length, incoming strength, service time and lane capacity. The four parameters are trained with four deep learning techniques, e.g., Long Short-Term Memory (LSTM), LSTM with Optuna, Gated Recurrent Unit (GRU), and GRU with Optuna. Real transaction data for this investigation was provided by the Expressway Authority of Thailand (EXAT), covering both weekdays and weekends and encompassing both manual toll collection (MTC) and electronic toll collection (ETC) systems. From the experiments, the GRU with Optuna model has shown to outperform the other models and previous frameworks. This indicates a significant advancement in toll plaza management and has the potential to lead to a more efficient toll collection system.