Capsules TCN Network for Urban Computing and Intelligence in Urban Traffic Prediction

Li, Dazhou; Lin, Chuan; Gao, Wei; Chen, Zeying; Wang, Zeshen; Liu, Guangqi

doi:10.1155/2020/6896579

Cited by 17 publications

(7 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Compared with the TCN module, because the normalization part is already included in the attention mechanism, the Weight norm layer and the ReLU layer in the residual block are deleted. Use FATCN to mine the potential relationship between the input features and obtain the weighted input sequence; then extract the hidden time sequence correlation information from the weighted input feature sequence at the ATGRU layer, and mine the relevant feature time sequence information and current time data through the time sequence attention layer [19]. And assign time attention weights to it to enhance the expressive ability of key historical moment information, obtain weighted comprehensive time series information status, and finally send it to the fully connected layer to output future closing price predictions [20].…”

Section: Fatcn-tagru Modelmentioning

confidence: 99%

Stock Price Prediction Model based on Dual Attention and TCN

Fu¹,

He²

2022

Signal, Image Processing and Embedded Systems Trends

View full text Add to dashboard Cite

The stock market is affected by many variables and factors, and the current forecasting models for time series are often difficult to capture the complex laws among multiple factors. Aiming at this problem, a stock price prediction model based on dual attention mechanism and temporal convolutional network is proposed. First, a convolution network more suitable for time series is used as the feature extraction layer. Feature attention is introduced to dynamically mine the potential correlation between the input factor features and closing prices. Second, based on Gated Recurrent Unit, on the other hand, a temporal attention mechanism is introduced to improve the model's ability to learn important time points and obtain importance measures from a temporal perspective. The experimental results show that the proposed model performs better than the traditional prediction model in the error index of stock price prediction and realizes the interpretability of the model in terms of index characteristics and time.

show abstract

Section: Fatcn-tagru Modelmentioning

confidence: 99%

Stock Price Prediction Model based on Dual Attention and TCN

Fu¹,

He²

2022

Signal, Image Processing and Embedded Systems Trends

View full text Add to dashboard Cite

show abstract

“…Zhang et al [4] proposed a short-term traffic flow forecasting model based on a convolutional neural network (CNN). Li et al [5] considered the spatial-temporal dependences and weather features as inputs and proposed a regional traffic flow forecasting model through a capsule network and time convolution network (TCN). Ma et al [6] also used CNN to extract features and predict short-term traffic flow.…”

Section: Literature Reviewmentioning

confidence: 99%

Demand Forecasting of Online Car-Hailing with Combining LSTM + Attention Approaches

Yan

et al. 2021

Electronics

View full text Add to dashboard Cite

The accurate prediction of online car-hailing demand plays an increasingly important role in real-time scheduling and dynamic pricing. Most studies have found that the demand of online car-hailing is highly correlated with both temporal and spatial distributions of journeys. However, the importance of temporal and spatial sequences is not distinguished in the context of seeking to improve prediction, when in actual fact different time series and space sequences have different impacts on the distribution of demand and supply for online car-hailing. In order to accurately predict the short-term demand of online car-hailing in different regions of a city, a combined attention-based LSTM (LSTM + Attention) model for forecasting was constructed by extracting temporal features, spatial features, and weather features. Significantly, an attention mechanism is used to distinguish the time series and space sequences of order data. The order data in Haikou city was collected as the training and testing datasets. Compared with other forecasting models (GBDT, BPNN, RNN, and single LSTM), the results show that the short-term demand forecasting model LSTM + Attention outperforms other models. The results verify that the proposed model can support advanced scheduling and dynamic pricing for online car-hailing.

show abstract

“…Shallow machine learning methods (e.g., support vector regression [37] and k-nearest neighbor [32]) can capture non-linear dependencies, but they need hand-crafted features, which require experts to do this. As the development of deep learning, [13,20,43] use simple univariate time series models such as RNNs and TCNs to capture temporal dependence of each sensor individually, ignoring the spatial correlations between sensors. Later, researchers use CNNs to extract the spatial dependence in the image-based traffic forecasting task [28,41,42], and CNNs are limited in the road network-based traffic forecasting task.…”

Section: Traffic Forecastingmentioning

confidence: 99%

CDGNet: A Cross-Time Dynamic Graph-based Deep Learning Model for Traffic Forecasting

Yu¹,

Qin²,

Luo³

et al. 2021

Preprint

View full text Add to dashboard Cite

Traffic forecasting is important in intelligent transportation systems of webs and beneficial to traffic safety, yet is very challenging because of the complex and dynamic spatio-temporal dependencies in real-world traffic systems. Prior methods use the pre-defined or learnable static graph to extract spatial correlations. However, the static graph-based methods fail to mine the evolution of the traffic network. Researchers subsequently generate the dynamic graph for each time slice to reflect the changes of spatial correlations, but they follow the paradigm of independently modeling spatio-temporal dependencies, ignoring the cross-time spatial influence. In this paper, we propose a novel cross-time dynamic graph-based deep learning model, named CDGNet, for traffic forecasting. The model is able to effectively capture the cross-time spatial dependence between each time slice and its historical time slices by utilizing the cross-time dynamic graph. Meanwhile, we design a gating mechanism to sparse the cross-time dynamic graph, which conforms to the sparse spatial correlations in the real world. Besides, we propose a novel encoder-decoder architecture to incorporate the cross-time dynamic graph-based GCN for multi-step traffic forecasting. Experimental results on three real-world public traffic datasets demonstrate that CDGNet outperforms the state-of-the-art baselines. We additionally provide a qualitative study to analyze the effectiveness of our architecture. CCS CONCEPTS• Information systems → Spatial-temporal systems; • Computing methodologies → Neural networks.

show abstract

Capsules TCN Network for Urban Computing and Intelligence in Urban Traffic Prediction

Cited by 17 publications

References 36 publications

Stock Price Prediction Model based on Dual Attention and TCN

Stock Price Prediction Model based on Dual Attention and TCN

Demand Forecasting of Online Car-Hailing with Combining LSTM + Attention Approaches

CDGNet: A Cross-Time Dynamic Graph-based Deep Learning Model for Traffic Forecasting

Contact Info

Product

Resources

About