Self-Organizing Traffic Flow Prediction with an Optimized Deep Belief Network for Internet of Vehicles

Area of Science:

• Computer Science
• Artificial Intelligence
• Transportation Engineering

Background:

• Fast network connectivity is crucial for broadcasting time-critical traffic information in Internet of Vehicles (IoV) and vehicular sensor networks (VSN).
• Accurate traffic prediction can significantly improve traffic congestion, operational efficiency, reduce commute times, noise, and carbon emissions.

Purpose of the Study:

• To develop a novel approach for predicting traffic flow volume using traffic data from self-organizing vehicular networks.
• To enhance the accuracy of traffic flow prediction models for better traffic management.

Main Methods:

• Utilized a Deep Belief Network (DBN), a probabilistic generative neural network with multiple Restricted Boltzmann Machine (RBM) auto-encoder layers.
• Employed time series data from roadside units (RSUs) across five highway links for training a three-layer DBN.
• Applied back-propagation for fine-tuning RBM weights and the Firefly Algorithm (FFA) for optimizing DBN topology and learning rate.
• Assessed prediction model accuracy using Monte Carlo simulations.

Main Results:

• The proposed DBN model demonstrated superior performance accuracy in predicting traffic flow compared to existing literature approaches.
• The model effectively extracts and learns key input features for traffic flow prediction.

Conclusions:

• The developed DBN-based approach offers a highly accurate method for traffic flow prediction in vehicular networks.
• This approach can effectively address traffic congestion issues and provide valuable guidance for road users and traffic regulators.