scispace - formally typeset
Open AccessProceedings Article

Deep Learning: A Generic Approach for Extreme Condition Traffic Forecasting.

TLDR
This work builds a deep neural network based on long short term memory (LSTM) units and applies Deep LSTM to forecast peak-hour traffic and manages to identify unique characteristics of the traffic data.
Abstract
Traffic forecasting is a vital part of intelligent transportation systems. It becomes particularly challenging due to short-term (e.g., accidents, constructions) and long-term (e.g., peak-hour, seasonal, weather) traffic patterns. While most of the previously proposed techniques focus on normal condition forecasting, a single framework for extreme condition traffic forecasting does not exist. To address this need, we propose to take a deep learning approach. We build a deep neural network based on long short term memory (LSTM) units. We apply Deep LSTM to forecast peak-hour traffic and manage to identify unique characteristics of the traffic data. We further improve the model for postaccident forecasting with Mixture Deep LSTM model. It jointly models the normal condition traffic and the pattern of accidents. We evaluate our model on a realworld large-scale traffic dataset in Los Angeles. When trained end-to-end with suitable regularization, our approach achieves 30%-50% improvement over baselines. We also demonstrate a novel technique to interpret the model with signal stimulation. We note interesting observations from the trained neural network.

read more

Citations
More filters
Book ChapterDOI

STCNet: Spatial-Temporal Convolution Network for Traffic Speed Prediction

TL;DR: Wang et al. as discussed by the authors proposed a Spatial-Temporal Convolution Network (STCNet), which mainly consists of a temporal block and a spatial block to model the short-term and long-term dependencies via different receptive fields.
Journal ArticleDOI

Pre-Trained Bidirectional Temporal Representation for Crowd Flows Prediction in Regular Region

TL;DR: Crowd flows prediction in regular gridded regions is studied and a model called Pre-trained Bidirectional Temporal Representation (PBTR) based on Transformer encoder is proposed capable of modeling very long-term temporal dependency automatically, which outperforms RNN-based methods and domain knowledge- based methods.
Journal ArticleDOI

Directional assessment of traffic flow extremes

TL;DR: In this article, a dimension reduction technique called principal component analysis (PCA) in an asymmetric norm is applied to assess the extremes of the traffic flow curves in a coherent way.
Posted Content

Error-feedback Stochastic Configuration Strategy on Convolutional Neural Networks for Time Series Forecasting

Xinze Zhang, +2 more
- 03 Feb 2020 - 
TL;DR: A novel Error-feedback Stochastic Configuration strategy to construct a random Convolutional Neural Network (ESC-CNN) for time series forecasting task, which builds the network architecture adaptively and exhibits strong predictive power in comparison to trained Convolution Neural Networks and Long Short-Term Memory models.
References
More filters
Posted Content

Sequence to Sequence Learning with Neural Networks

TL;DR: This paper presents a general end-to-end approach to sequence learning that makes minimal assumptions on the sequence structure, and finds that reversing the order of the words in all source sentences improved the LSTM's performance markedly, because doing so introduced many short term dependencies between the source and the target sentence which made the optimization problem easier.
Posted Content

Show, Attend and Tell: Neural Image Caption Generation with Visual Attention

TL;DR: This paper proposed an attention-based model that automatically learns to describe the content of images by focusing on salient objects while generating corresponding words in the output sequence, which achieved state-of-the-art performance on three benchmark datasets: Flickr8k, Flickr30k and MS COCO.
Posted Content

Recurrent Neural Network Regularization

TL;DR: This paper shows how to correctly apply dropout to LSTMs, and shows that it substantially reduces overfitting on a variety of tasks.
Journal ArticleDOI

Traffic Flow Prediction With Big Data: A Deep Learning Approach

TL;DR: A novel deep-learning-based traffic flow prediction method is proposed, which considers the spatial and temporal correlations inherently and is applied for the first time that a deep architecture model is applied using autoencoders as building blocks to represent traffic flow features for prediction.
Posted Content

Practical recommendations for gradient-based training of deep architectures

TL;DR: Overall, this chapter describes elements of the practice used to successfully and efficiently train and debug large-scale and often deep multi-layer neural networks and closes with open questions about the training difficulties observed with deeper architectures.
Related Papers (5)