Loading...

Traffic Prediction using Graph Neural Network

Sadeghi, Danial | 2023

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58516 (46)
  4. University: Sharif University of Technology
  5. Department: Energy Engineering
  6. Advisor(s): Boroushaki, Mehrdad
  7. Abstract:

  8. چکیده (انگلیسی) Prediction of traffic parameters is one of the most important needs in the field of intelligent transportation. Traffic forecasting allows people to choose the best times and routes to travel using various data such as date and time, events and weather conditions. Traffic forecasting helps reduce traffic in cities and save energy. This information is important for urban management; Because it provides the possibility of improving public transportation systems and traffic measures, and in this way, it is possible to propose the best solutions for managing traffic and increasing the quality of life of citizens. Although various parameters such as traffic speed, traffic flow, traffic density, etc. are used in traffic forecasting; But in this research, we only use the traffic speed parameter. Doing this comes with many challenges due to the complexity of traffic data. One of these challenges is to consider the spatial dependencies and the effect of the passage of a street and intersection on the surrounding streets and intersections, which is proposed to solve this challenge of neural graph networks. In this research, two different architectures of graph neural networks, namely STGCN model and DCRNN model, have been used to predict the traffic speed parameter of Los Angeles city. STGCN model is a deep learning model for traffic prediction in space and time. This model uses convolutional graph layers along with spatial and temporal information to model complex traffic patterns. According to the features of the neural network of the graph, this model can also consider the spatial connections between different points in the traffic network and thereby increase the accuracy of the predictions. DCRNN model is a deep learning model for traffic prediction in road networks. This model uses convolutional graph layers and recurrent networks to understand spatial and temporal traffic patterns. By applying graph techniques, DCRNN is able to model the connections between different points in the road network and predict traffic changes over time. In the DCRNN model, diffusion is a process that transfers information from one point to adjacent points in the graph. By applying diffusion, the model is able to understand the effect of changes in a specific point on the points around it and in general on the traffic in the network. This feature helps the model to consider better spatial information in traffic prediction and perform better. Then, with the help of reinforcement learning, we created an ensemble model in order to improve the results of these two models, and finally we came to the conclusion that the accuracy of a separate DCRNN model is higher than a separate STGCN model, and the ensemble model of these two models is the most accurate
  9. Keywords:
  10. Traffic ; Reinforcement Learning ; Deep Learning ; Artificial Intelligence ; Intelligent Transportation System (ITS) ; Graph Neural Network ; Traffic Prediction

 Digital Object List

 Bookmark

No TOC