Long-term Traffic Prediction Research Articles

STANN: A Spatio–Temporal Attentive Neural Network for Traffic Prediction

Recently, traffic prediction based on deep learning methods has attracted much attention. However, there still exist two major challenges, namely, dynamic spatio-temporal dependences among network-wide links and long-term traffic prediction for the next few hours. To address these two challenges, this paper proposes a spatio-temporal attentive neural network (STANN) for the network-wide and long-term traffic prediction. The STANN captures the spatial–temporal dependences based on the encoder–decoder architecture with the attention mechanisms. In the encoder, the STANN learns the spatio-temporal dependences from historical traffic series using a recurrent neural network (RNN) with long short-term memory (LSTM) units, in which a new spatial attention model is developed to consider the contribution of each link to the network-wide prediction. In the decoder, the STANN exploits another RNN with LSTM units and a temporal attention model to select the relevant and important historical spatio-temporal dependences from the encoder for long-term traffic prediction. Finally, we conduct extensive experiments to evaluate STANN on three real-world traffic datasets. The experimental results show that the STANN is significantly better than other state-of-the-art models.

Traffic Flow Data Prediction Using Residual Deconvolution Based Deep Generative Network

Traffic flow prediction is quite crucial for estimating the future traffic states, efficient and accurate prediction models greatly contribute to the smooth traffic of road networks. However, existing methods mainly concentrate on short-term prediction. The challenging task of long-term flow prediction for the next day, as the important reference of traffic management, is still not well solved. In this paper, we present a residual deconvolution based deep generative network (RDBDGN) to handle the problem of long-term traffic flow prediction. The proposed method consists of a generator and a discriminator. The generator is composed of multi-channel residual deconvolutional neural networks, and the discriminator contains a convolutional neural network which aims to optimize the adversarial training process. The experiments are evaluated based on the traffic flow data of elevated highways, presented results demonstrate that our approach outperforms the state-of-the-art works.

Spatiotemporal Recurrent Convolutional Networks for Traffic Prediction in Transportation Networks.

Predicting large-scale transportation network traffic has become an important and challenging topic in recent decades. Inspired by the domain knowledge of motion prediction, in which the future motion of an object can be predicted based on previous scenes, we propose a network grid representation method that can retain the fine-scale structure of a transportation network. Network-wide traffic speeds are converted into a series of static images and input into a novel deep architecture, namely, spatiotemporal recurrent convolutional networks (SRCNs), for traffic forecasting. The proposed SRCNs inherit the advantages of deep convolutional neural networks (DCNNs) and long short-term memory (LSTM) neural networks. The spatial dependencies of network-wide traffic can be captured by DCNNs, and the temporal dynamics can be learned by LSTMs. An experiment on a Beijing transportation network with 278 links demonstrates that SRCNs outperform other deep learning-based algorithms in both short-term and long-term traffic prediction.

How long will the traffic flow time series keep efficacious to forecast the future?

Abstract This paper investigate how long will the historical traffic flow time series keep efficacious to forecast the future. In this frame, we collect the traffic flow time series data with different granularity at first. Then, using the modified rescaled range analysis method, we analyze the long memory property of the traffic flow time series by computing the Hurst exponent. We calculate the long-term memory cycle and test its significance. We also compare it with the maximum Lyapunov exponent method result. Our results show that both of the freeway traffic flow time series and the ground way traffic flow time series demonstrate positively correlated trend (have long-term memory property), both of their memory cycle are about 30 h. We think this study is useful for the short-term or long-term traffic flow prediction and management.

Estimating Mobile Traffic Demand Using Twitter

In this letter, the authors show that structured social media data can act as an accurate predictor for wireless data demand patterns at a high spatial-temporal resolution. A case-study is performed on Greater London covering a 5000 km $^{2}$ area. The data used includes over 0.6 million geo-tagged Twitter data, over 1 million mobile phone data demand records, and U.K. census data. The analysis shows that social media activity (Tweets/s ${n}$ ) can accurately predict the long-term traffic demand for both the uplink and downlink channels. The relationship between social media activity and traffic demand obeys a power law and the model explains for over 71%–79% of the variance in real traffic demand. This is a significant improvement over existing methods of long-term traffic prediction such as census population data ( $ {R^{2}=0.57}$ ). The authors also show that social media data can also forward predict short-term traffic demand for up to 2 h on the same day and for the same time in the following 2–3 days.

Repeatability and Similarity of Freeway Traffic Flow and Long-Term Prediction Under Big Data

In this paper, by splitting a traffic flow series into basis series and deviation series, the concepts of similarity and repeatability of traffic flow patterns are defined using the statistic average values of the basis series and the deviation series and are further verified through the real-time big traffic data of 82 days with a sampling period of 5 min collected from two typical ones among a total of 102 detecting sites in Shenzhen, China. Meanwhile, based on the repeatability and the similarity of the traffic flow series, a novel long-term forecasting method for traffic flow is developed, and hybrid forecasting algorithms for short-/long-term traffic flow prediction are also proposed. The effectiveness of these algorithms is verified by using the real-time data.

Network traffic prediction based on wavelet FARIMA model

Many researches indicate that the Internet traffic exhibits Long-Range Dependence(LRD) and Short-Range Dependence(SRD) features simultaneously.To capture these features of traffic flow,a prediction method was proposed based on W-FARIMA(Wavelet Fractal Autoregressive Integrated Moving Average) model.First,original data series were decomposed by Haar wavelet into a low frequency signal and several high frequency signals.Second,the low frequency signal was modeled and predicted by FARIMA model.Third,the high frequency signals were predicted based on the weighted first order local prediction method.Finally,the respective prediction series were synthesized to get the final prediction data.The experimental results and mathematical analysis confirm that the model performs well in the long-term network traffic prediction.

3-D Physical Motion-Based Bandwidth Prediction for Video Conferencing

Interactive video services such as video conferencing, distance learning, and online video games over the Internet and wireless networks are becoming increasingly prevalent. Because of limited bandwidth on networks and the bandwidth-hungry nature of video, interactive video requires extremely efficient resource management to gain real-time performance. Unlike pregenerated video whose traffic profiles can be computed in advance, efficiency and accuracy of dynamic resource allocation methods for interactive video depend critically on the performance of traffic prediction. Using either traffic data or image features, existing traffic prediction schemes can only provide a short-term traffic prediction. Based on a three-dimensional object's motion, this paper presents a new bandwidth prediction approach for video conferencing. We show that there is a strong correlation between video conferencing traffic and real motion of objects. The real motion can be predicted by the powerful technique Kalman filtering, and the estimated motion is used to make a long-term bandwidth prediction. The new traffic prediction model is tested and compared with the frame-based adaptive normalized least mean square error linear predictor and optical flow-based method with Kalman filtering. Experimental results show that our proposed traffic prediction model achieves much higher accuracy in long-term traffic prediction, which provides the possibility for networks to allocate resources efficiently for video conferencing services.

Real-Time VBR Video Traffic Prediction for Dynamic Bandwidth Allocation

In this paper, we systematically investigate the long-term, online, real-time variable-bit-rate (VBR) video traffic prediction, which is the key and complicated component for advanced predictive dynamic bandwidth control and allocation framework for the future networks and Internet multimedia services. We focus on neural network-based approach for traffic prediction and demonstrate that the prediction performance and robustness of neural network predictors can be significantly improved through multiresolution learning. We show that neural network traffic predictor trained through the multiresolution learning (called multiresolution learning NN traffic predictor) can successfully predict various real-world VBR video traffic up to hundreds of frames in advance, which then lays a solid foundation for predictive dynamic bandwidth control and allocation mechanism. Also, dynamic bandwidth control/allocation based on long-term traffic prediction is discussed in detail.