Learning Traffic as Images: A Deep Convolutional Neural Network for Large-Scale Transportation Network Speed Prediction

Abstract

This paper proposes a convolutional neural network (CNN)-based method that learns traffic as images and predicts large-scale, network-wide traffic speed with a high accuracy. Spatiotemporal traffic dynamics are converted to images describing the time and space relations of traffic flow via a two-dimensional time-space matrix. A CNN is applied to the image following two consecutive steps: abstract traffic feature extraction and network-wide traffic speed prediction. The effectiveness of the proposed method is evaluated by taking two real-world transportation networks, the second ring road and north-east transportation network in Beijing, as examples, and comparing the method with four prevailing algorithms, namely, ordinary least squares, k-nearest neighbors, artificial neural network, and random forest, and three deep learning architectures, namely, stacked autoencoder, recurrent neural network, and long-short-term memory network. The results show that the proposed method outperforms other algorithms by an average accuracy improvement of 42.91% within an acceptable execution time. The CNN can train the model in a reasonable time and, thus, is suitable for large-scale transportation networks.

Road2Vec: Measuring Traffic Interactions in Urban Road System from Massive Travel Routes

“…Existing traffic forecasting methods using deep learning architectures in recent years claim that their approaches can take the spatio-temporal correlations in road systems into account, however, there is no process for establishing spatial relationship in their methods [3], or only roads with linear geometric relationships on straight lines are considered [2]. In subsequent research, we will attempt to integrate the Road2Vec method with a deep learning architecture to further improve traffic forecasting performance.…”

“…(1) if more than one consecutive GPS points are mapped onto the same road segment, then the road segment is only counted once in the travel route; (2) if two consecutive GPS points are mapped onto different road segments that are not topologically adjacent in a road network, then we use the shortest path between the two road segments to form the travel route.…”

“…Different terms, such as spatial dependency/relationship in traffic [1,2] and spatial correlation [3][4][5][6], are used in the literature to express such relationship between neighboring roads. In this paper, we use the term traffic interaction to describe the traffic influence between neighboring roads, which is fundamentally caused by the dynamic vehicle movements from one road to another.…”

Cluster-Based LSTM Network for Short-Term Passenger Flow Forecasting in Urban Rail Transit

“…The second category is the convolutional-neuralnetwork (CNN)-based models, which can extract spatial dependencies even when stations are far away from each other. CNN-based models always treat passenger flows as images so that the convolution operation can be conducted [11]. Residual network (ResNet) [12] is a typical framework using skip-connection between CNN layers and has been proved to be effective in STPFF such as spatiotemporal ResNet models [13,14].…”

Spatiotemporal Traffic State Prediction Based on Discriminatively Pre-trained Deep Neural Networks

“…Lv et al used a stacked autoencoder model to learn features that capture the nonlinear spatial and temporal correlations from the traffic data, then forwarded these features to the output layer to predict the traffic flow [17]. In an another recent study by Ma et al, spatiotemporal speed matrices were considered as images and used to train a convolution neural network, and then this network was used to predict large-scale, network-wide traffic speed with high accuracy [18].…”