Traffic flow prediction using Deep Sedenion Networks

Alabi Bojesomo,Panos Liatsis,Hasan Al Marzouqi

DOI: https://doi.org/10.48550/arXiv.2012.03874

2020-12-11

Abstract:In this paper, we present our solution to the Traffic4cast2020 traffic prediction challenge. In this competition, participants are to predict future traffic parameters (speed and volume) in three different cities: Berlin, Istanbul and Moscow. The information provided includes nine channels where the first eight represent the speed and volume for four different direction of traffic (NE, NW, SE and SW), while the last channel is used to indicate presence of traffic incidents. The expected output should have the first 8 channels of the input at six future timing intervals (5, 10, 15, 30, 45, and 60min), while a one hour duration of past traffic data, in 5mins intervals, are provided as input. We solve the problem using a novel sedenion U-Net neural network. Sedenion networks provide the means for efficient encoding of correlated multimodal datasets. We use 12 of the 15 sedenion imaginary parts for the dynamic inputs and the real sedenion component is used for the static input. The sedenion output of the network is used to represent the multimodal traffic predictions. Proposed system achieved a validation MSE of 1.33e-3 and a test MSE of 1.31e-3.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper attempts to address the problem of predicting future traffic parameters (speed and flow) in three different cities (Berlin, Istanbul, and Moscow). Specifically, given the traffic data from the past 1 hour (with a time point every 5 minutes), it is necessary to predict the traffic conditions for the next six time intervals (5 minutes, 10 minutes, 15 minutes, 30 minutes, 45 minutes, and 60 minutes). The paper proposes a solution based on Deep Sedenion Networks, using a novel Sedenion U-Net neural network to efficiently encode and process multimodal data. ### Main Issues 1. **Predicting Future Traffic Parameters**: Predicting traffic speed and flow for the next six time intervals based on the traffic data from the past 1 hour. 2. **Handling Multimodal Data**: The input data includes dynamic traffic information (speed and flow in different directions) and static traffic information (such as the presence of traffic events). ### Solution 1. **Sedenion U-Net Neural Network**: Using sedenions (a type of 16-dimensional hypercomplex number) to efficiently represent and process multimodal data. 2. **Input Representation**: Static inputs are used as the real part of the sedenion, while dynamic inputs are used as the 12 imaginary parts of the sedenion. 3. **Output Representation**: The prediction results are extracted from the last layer of the network, corresponding to the traffic parameters for the next six time intervals. 4. **Model Structure**: The model includes feature extraction blocks, encoding blocks, decoding blocks, etc., with most convolutional layers being sedenion convolutional layers. ### Experimental Results - **Validation Set Mean Squared Error (MSE)**: 1.33e-3 - **Test Set Mean Squared Error (MSE)**: 1.31e-3 Through these methods, the paper demonstrates how to efficiently handle and predict multimodal traffic data using the Sedenion U-Net neural network.

Traffic flow prediction using Deep Sedenion Networks

Spatial–Temporal Deep Tensor Neural Networks for Large-Scale Urban Network Speed Prediction

Daily Long-Term Traffic Flow Forecasting Based on a Deep Neural Network

A New Hybrid Deep Learning Algorithm for Prediction of Wide Traffic Congestion in Smart Cities

Traffic map prediction using UNet based deep convolutional neural network

TrafficPro: A Road Speed Prediction Framework for Urban Signal-Controlled Road Networks

Deep Autoencoder Neural Networks for Short-Term Traffic Congestion Prediction of Transportation Networks

Traffic4cast at NeurIPS 2022 -- Predict Dynamics along Graph Edges from Sparse Node Data: Whole City Traffic and ETA from Stationary Vehicle Detectors

Traffic4cast -- Large-scale Traffic Prediction using 3DResNet and Sparse-UNet

A Deep Ensemble Approach for Long-Term Traffic Flow Prediction

Large-Scale Road Network Traffic Congestion Prediction Based on Recurrent High-Resolution Network

The Prediction of Urban Road Traffic Congestion by Using a Deep Stacked Long Short-Term Memory Network

Utilizing UNet for the future traffic map prediction task Traffic4cast challenge 2020

Urban Traffic Flow Prediction with Deep Neural Network

A Novel Deep Stacking-Based Ensemble Approach for Short-Term Traffic Speed Prediction

Traffic flow prediction by an ensemble framework with data denoising and deep learning model

A deep-learning model for urban traffic flow prediction with traffic events mined from twitter

Towards Good Practices of U-Net for Traffic Forecasting

Short-Term Traffic Flow Prediction via Improved Mode Decomposition and Self-Attention Mechanism Based Deep Learning Approach

Traffic congestion anomaly detection and prediction using deep learning

Motorway Traffic Flow Prediction using Advanced Deep Learning