A Deep Learning Algorithm for Travel Time Prediction

Abstract

Accurate prediction of vehicle travel times is crucial for enhancing intelligent transportation systems, optimizing routing solutions, improving ride-sharing services, and managing traffic effectively. There are various methods available for predicting vehicle travel times between two locations, including both model-based and data-driven approaches. Traditional models often fall short because they assume Euclidean distance when predicting travel times between points. In this study, we focus on predicting vehicle travel times for road segments and entire routes using detailed trajectory data that includes latitude, longitude, time of day, time of week, driver habits, and driver ID. Each trajectory consists of a sequence of GPS points that track a vehicle's movements over time. By defining a road segment as the route between three consecutive GPS points, we can break down the trajectory into smaller segments, enabling more accurate travel time estimates. Given the complexity of travel time prediction, which is influenced by traffic flow conditions at different times and locations, we propose a deep learning algorithm. This algorithm utilizes advanced techniques, including Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Temporal Convolutional Networks (TCNs). Our approach demonstrates significant improvements over existing methods. Using the Mean Absolute Percent Error (MAPE) metric, we compared our model with established ones, employing large-scale Chengdu taxi datasets. Our results indicate a 2.9% improvement in travel time prediction accuracy, highlighting our model's potential to surpass current solutions and paving the way for future research in travel time estimation.

Keywords:

• Travel time prediction Keyword: Travel time prediction
• Trajectory data Keyword: Trajectory data
• Deep learning Keyword: Deep learning
• Long short-term memory networks Keyword: Long short-term memory networks
• Convolutional neural networks Keyword: Convolutional neural networks
• Temporal convolutional networks Keyword: Temporal convolutional networks

How to Cite: