A Multi-Graph Convolutional Neural Network Model for Short-Term Prediction of Turning Movements at Signalized Intersections

TL;DR

This paper introduces a multigraph convolutional neural network (MGCNN) for short-term prediction of turning movements at signalized intersections, effectively capturing spatial and temporal traffic data variations.

Contribution

The study presents a novel deep learning architecture combining multigraph structures with spectral convolution for improved traffic movement prediction.

Findings

MGCNN outperforms baseline models in prediction accuracy.

The model achieves a mean squared error of 0.9.

Effective modeling of spatial-temporal traffic data variations.

Abstract

Traffic flow forecasting is a crucial first step in intelligent and proactive traffic management. Traffic flow parameters are volatile and uncertain, making traffic flow forecasting a difficult task if the appropriate forecasting model is not used. Additionally, the non-Euclidean data structure of traffic flow parameters is challenging to analyze from both spatial and temporal perspectives. State-of-the-art deep learning approaches use pure convolution, recurrent neural networks, and hybrid methods to achieve this objective efficiently. However, many of the approaches in the literature rely on complex architectures that can be difficult to train. This complexity also adds to the black-box nature of deep learning. This study introduces a novel deep learning architecture, referred to as the multigraph convolution neural network (MGCNN), for turning movement prediction at intersections. The proposed architecture combines a multigraph structure, built to model temporal variations in traffic data, with a spectral convolution operation to support modeling the spatial variations in traffic data over the graphs. The proposed model was tested using twenty days of flow and traffic control data collected from an arterial in downtown Chattanooga, TN, with ten signalized intersections. The model's ability to perform short-term predictions over 1, 2, 3, 4, and 5 minutes into the future was evaluated against four baseline state-of-the-art models. The results showed that our proposed model is superior to the other baseline models in predicting turning movements with a mean squared error (MSE) of 0.9