Road Traffic Speed Prediction Using Graph Convolutional Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting road traffic speed, such as temporal and spatial-temporal approaches, face challenges with high computational complexity and inaccurate predictions due to their inability to effectively capture spatial dependencies in road networks.
Innovation Solution
A system that uses a recurrent neural network to process raw trajectory data, extract node and edge features, and perform graph convolutions with learnable parameters to predict road traffic speed, reducing computational complexity and improving accuracy by directly capturing spatial dependencies through node and edge embedding layers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If spatial-temporal approaches with multiple graph convolutions are used to capture spatial dependencies, then prediction accuracy improves, but computational complexity increases
Solution Approach 1:
The patent extracts and utilizes pre-computed topological features (node2vec embeddings, betweenness centrality, degree, clustering coefficient) from the road network graph structure. By extracting these spatial dependency features directly from the graph topology rather than through multiple iterative convolutions, the method achieves accurate spatial dependency capture with reduced computational complexity
Solution Approach 2:
The patent performs preliminary computation of topological features and node embeddings before the main prediction process. The node2vec embeddings and graph theoretical metrics are pre-computed and stored, allowing the prediction model to directly utilize these features without performing multiple graph convolution operations during inference, thus reducing runtime complexity
2Reliability
If temporal approaches are used to capture sequential characteristics, then daily and periodic trends are captured, but spatial correlations in the network are not accounted for
Solution Approach 1:
The patent merges temporal approaches (LSTM for sequential processing) with spatial feature extraction (node2vec embeddings, graph theoretical metrics) into a unified model. The input to the LSTM includes both historical speed data and pre-computed spatial features, allowing the model to simultaneously capture temporal dependencies and spatial correlations without losing either type of information
Data Source
AI summary
A system for predicting road speed traffic is disclosed. The system may be configured to receive and process raw trajectory data to determine processed trajectory data; obtain node features representing information about road segment characteristics; obtain edge features representing information about interactions between the node features; determine a learned graph representation of a road network based on a node embedding of the node features and an edge embedding of the edge features; determine at least one hidden states value based on a graph convolution of the learned graph representation through the at least one encoder neural network; and predict road speed traffic based on the at least one hidden states value through at least one decoder neural network.