Distributed Traffic Prediction Models for Privacy-Safe Route Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing route planning methods for vehicular traffic systems are inadequate due to the dynamic nature of traffic conditions, reliance on potentially misleading crowdsourced information, and limitations in centralized and on-board learning, which fail to provide accurate short-term, medium-term, and long-term traffic predictions, leading to suboptimal route planning and increased congestion.
Innovation Solution
A distributed machine learning approach using federated learning and multi-task learning techniques, where infrastructure devices act as learning servers and data collectors as agents, train localized traffic models based on clustered data to create robust global traffic models, considering location, time, and weather factors, and predict traffic with multi-horizon time domains, while maintaining data privacy and reducing communication overhead.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If centralized learning at cloud is used to collect sufficient data for training machine learning algorithms, then data completeness for training is improved, but communication bandwidth requirements increase and data privacy/security issues worsen
Solution Approach 1:
The patent segments the centralized learning system into distributed edge learning nodes (vehicles and infrastructure devices) that independently train local models. Each node processes data locally rather than centralizing all data at the cloud, thereby reducing communication bandwidth requirements while maintaining model training effectiveness through federated learning aggregation.
Solution Approach 2:
The patent introduces federated learning as an intermediary mechanism that enables collaborative model training without direct data sharing. The global model serves as an intermediary that aggregates knowledge from multiple local models through parameter sharing, eliminating the need for raw data transmission while achieving comprehensive learning.
2Measurement precision
If machine learning algorithms are trained using large amount of different data types for accurate traffic prediction, then prediction accuracy is improved, but data privacy and security risks worsen
Solution Approach 1:
The patent extracts only the essential model parameters and gradients needed for training from the raw traffic data, leaving the sensitive raw data localized at each edge device. This extraction approach enables accurate model training while eliminating the privacy risks associated with centralized raw data storage and transmission.
Solution Approach 2:
The patent uses model parameter copies instead of original data copies for sharing between nodes. Each vehicle and infrastructure device maintains local copies of the global model parameters and contributes updated parameters through federated learning, achieving collaborative training without exposing sensitive raw traffic data.
3Area of stationary object
If existing route planning methods rely on crowdsourcing information from multiple vehicles, then route planning coverage is improved, but traffic density estimation accuracy worsens due to misleading decisions
Solution Approach 1:
The patent implements a feedback mechanism where vehicles receive predicted traffic information from the distributed learning model and use this information to make route decisions. The system continuously updates predictions based on actual traffic conditions observed by edge devices, creating a closed-loop feedback system that improves both coverage and accuracy over time.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
A distributed machine learning based traffic prediction method is provided for predicting traffic of roads. In this case, the distributed machine learning based traffic prediction method includes distributing global multi-task traffic models by a learning server to learning agents to locally train the traffic models, uploading locally trained traffic models by learning agents to the learning server, updating global multi-task traffic models by the learning server using locally trained traffic model parameters acquired from learning agents, generating a time-dependent global traffic map by the learning server using the well trained global multi-task traffic models, distributing the time dependent global traffic map to vehicles traveling on the roads, and computing an optimal travel route with the least travel time by a vehicle using the time-dependent global traffic map based on a driving plan.