Traffic Prediction Model Using Graph Convolutional Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current traffic prediction techniques based on machine learning fail to accurately capture spatial and temporal relationships in traffic data, leading to poor predictions and require extensive and costly training processes.
Innovation Solution
A computer-implemented method using a learning system with a convolution engine and encoder-decoder to construct a graph representation of localities, convolve traffic data with neighboring data, and encode/decode it into predicted traffic data, allowing for efficient training and knowledge transfer across localities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If state of the art machine learning techniques are used for traffic prediction, then the system can process traffic data, but the spatial and temporal relationships are not captured accurately resulting in poor predictions
Solution Approach 1:
The patent segments the traffic prediction problem into distinct spatial and temporal components. It uses graph convolutional networks to handle spatial relationships between road segments and recurrent neural networks to handle temporal dependencies, allowing each aspect to be processed with specialized architecture that captures its unique characteristics accurately
Solution Approach 2:
The patent transforms the traditional flat traffic data structure into a multi-dimensional representation by constructing graph structures that encode spatial relationships and temporal sequences that preserve temporal relationships. This dimensional transformation enables the model to capture complex spatio-temporal patterns that flat architectures cannot represent
2Measurement precision
If state of the art machine learning systems are trained with vast amounts of training data, then more comprehensive traffic patterns can be learned, but the training process becomes very tedious and costly
Solution Approach 1:
The patent performs preliminary action by pre-processing traffic data into graph representations with embedded spatial and temporal features before training. This preprocessing step organizes data in a structure that the model can efficiently learn from, reducing the amount of training data and time needed compared to training from raw data
Solution Approach 2:
The patent changes parameters by transforming traffic data into graph-based representations with specific structural properties. This parameter transformation allows the model to learn from fewer samples because the graph structure encodes prior knowledge about road networks and traffic flow patterns, reducing the need for extensive training data
3Device complexity
If traditional machine learning models are used, then the system architecture is simpler, but they fail to capture the actual physical characteristics of the road network
Solution Approach 1:
The patent introduces graph structures as an intermediary between raw traffic data and the neural network model. These graphs serve as a mediator that encodes the physical characteristics of road networks, allowing the model to learn from data while preserving the underlying spatial relationships without requiring complex model architecture
Data Source
AI summary
A computer implemented method for training a learning model for traffic prediction at respective localities by means of a learning system includes a convolution engine and an encoder-decoder. The method involves constructing a graph representation of the localities based on a spatial relation between the respective localities; populating the constructed graph with traffic data characterizing the traffic in the respective localities at respective time periods; convolving, by the convolution engine, for a respective locality and for a respective time period, the traffic data in the respective locality with the traffic data in its neighboring localities, thereby obtaining relation-based traffic representation; processing, by the encoder-decoder, for a respective locality and for a respective time period, the relation-based traffic representation, thereby obtaining a gradient information; and updating, for a respective locality, the learning model with the obtained gradient information, thereby training the learning model.