دل ارا قديري

Short-term travel time prediction has long been an important issue in transportation planning and an integral part of intelligent transportation systems. Nowadays, deep learning models address some of the deficiencies of previous methods and predict more accurately by considering the variables affecting traffic flow. Various methods exist for collecting the data needed for prediction as input to the models. Among them, real-time data from routing applications and GPS embedded in mobile phones and vehicles are more cost-effective and convenient for obtaining real-time data. Moreover, among different prediction models, efficient models capable of capturing patterns in historical data are chosen based on the research goal and data type to achieve more accurate results. Since speed and travel time are interrelated, average speed over a segment can be derived using travel time data and vice versa. Prediction accuracy depends on the future time intervals for which predictions are made; therefore, these intervals should be within an acceptable range for short-term prediction, typically between 5 to 60 minutes into the future. In this research, considering previous studies, artificial intelligence-based methods, including machine learning and deep learning, using LSTM, CNN, GRU, and Ensemble learning models, were used to predict travel time in short-term intervals of 5, 10, 15, 30, and 60 minutes. These models were implemented on highway segments, including 59 different segments from 5 highways in Tehran. By comparing the eva‎luation metrics results of each model, the GRU model provided more acceptable and accurate RMSE, MAE, and R² values and was selected as the superior model. It was also observed that prediction accuracy is higher for short-term intervals up to 30 minutes into the future, and after this interval, the model's accuracy decreases.

پيش‌بيني كوتاه‌مدت زمان سفر , هوش مصنوعي , مدلهاي يادگيري عميق , داده‌هاي زمان سفر , پيش بيني جريان ترافيك

Short-term Travel time Prediction , Artificial Intelligence , Deep Learning Models , Travel Time Data , Traffic Flow Forecasting

delara ghadiri

reza golshan khavas