Multi-mode Dynamic Residual Graph Convolution Network for Traffic Flow Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing traffic flow forecasting methods struggle to effectively capture and integrate the complex characteristics of different traffic modes in urban networks, which are influenced by various factors such as topology, weather, and emergencies, leading to inaccurate congestion predictions.
Innovation Solution
A traffic flow forecasting method based on a multi-mode dynamic residual graph convolution network is developed, utilizing two adjacency matrices to capture space and time dependencies, and a dynamic residual fusion mechanism to combine historical data with extracted features for improved forecasting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single adjacency matrix is used to capture traffic characteristics, then the model structure is simple, but it cannot capture traffic characteristics corresponding to different traffic modes
Solution Approach 1:
The patent divides the single adjacency matrix into two separate adjacency matrices: a first adjacency matrix for capturing space dependence characteristics and a second adjacency matrix for capturing time dependence characteristics. This segmentation allows each matrix to specialize in capturing specific traffic mode characteristics, thereby improving the model's adaptability to different traffic modes while maintaining manageable structural complexity through modular design.
2Measurement precision
If multiple adjacency matrices are used to capture different traffic modes, then the capture accuracy of traffic characteristics improves, but the computational complexity increases
Solution Approach 1:
The patent merges the outputs of the two adjacency matrices through a fusion mechanism that combines space dependence and time dependence characteristics. By integrating these two matrices' results, the model achieves comprehensive capture of different traffic modes' characteristics while managing computational complexity through efficient fusion operations rather than requiring separate processing for each matrix.
Solution Approach 2:
The patent designs a unified graph convolution network framework that can process both space and time dependence characteristics through the two adjacency matrices. This multi-functional design allows the same network architecture to handle different traffic modes (space-based and time-based) without requiring completely separate models, thereby improving measurement precision while controlling computational complexity through shared computational resources.
3Measurement precision
If historical data is not effectively integrated, then the forecasting model is simpler, but the forecasting accuracy is reduced
Solution Approach 1:
The patent performs preliminary action by extracting and pre-processing historical data features before feeding them into the forecasting model. The historical data is processed to extract relevant patterns and characteristics in advance, which are then integrated into the graph convolution network. This preliminary processing improves forecasting accuracy by ensuring the model receives well-processed historical information while reducing the complexity of real-time data integration during forecasting.
Data Source
AI summary
Disclosed is a traffic flow forecasting method based on a multi-mode dynamic residual graph convolution network, including following steps: constructing a relationship matrix and an adaptive matrix to learn the site dependence relationship for historical traffic data of traffic stations; using multi-mode dynamic graph convolution to extract traffic characteristics corresponding to different traffic modes; embedding the graph convolution into the gated cyclic neural network to realize the combination of space dependence and time dependence of traffic flow; connecting the network by using the dynamic residual, and combining the input traffic data with the decoding data to obtain the final forecasting value. The application utilizes two different methods to construct adjacency matrix, effectively captures traffic flow characteristics corresponding to different traffic modes, and dynamically fuses traffic flow characteristics of two different modes.