Article
Computer Science, Information Systems
Sonu Mathew, Srinivas S. Pulugurtha, Chaitanya Bhure, Sarvani Duvvuri
Summary: The focus of this research is to develop a robust model to accurately estimate the daily average traffic volume of all local roads. The 1D-CNN model, combined with domain knowledge of local road characteristics, was used to estimate the traffic volume. Comparison with other models showed that the 1D-CNN model outperformed the others.
Article
Chemistry, Multidisciplinary
Renyi Chen, Huaxiong Yao
Summary: This paper proposes a Hybrid Graph Model (HGM) for accurate traffic prediction, which constructs a static graph and a dynamic graph to represent the topological information of the traffic network and extract complex spatial-temporal features. The HGM combines graph neural network, convolutional neural network, and attention mechanism to improve prediction performance. Extensive experiments show that the HGM outperforms comparable state-of-the-art methods.
APPLIED SCIENCES-BASEL
(2023)
Article
Chemistry, Multidisciplinary
Mustapha Mukhtar, Ariyo Oluwasanmi, Nasser Yimen, Zhang Qinxiu, Chiagoziem C. Ukwuoma, Benjamin Ezurike, Olusola Bamisile
Summary: This study develops two novel hybrid neural network models for accurate prediction of global solar radiation. Compared with traditional artificial neural network models, the hybrid models show better performance in different countries across Africa. The results of this study are of great significance for finding more accurate methods of solar radiation estimation.
APPLIED SCIENCES-BASEL
(2022)
Article
Engineering, Civil
Kan Guo, Yongli Hu, Zhen Qian, Hao Liu, Ke Zhang, Yanfeng Sun, Junbin Gao, Baocai Yin
Summary: This paper introduces an optimized graph convolution recurrent neural network for traffic prediction, which can better explore the spatial and temporal information of traffic data and learns an optimized graph through a data-driven approach to reveal the latent relationship among road segments.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2021)
Article
Computer Science, Software Engineering
Gleb Tkachev, Steffen Frey, Thomas Ertl
Summary: The proposed machine learning approach detects and visualizes complex behavior in spatiotemporal volumes by training models to predict future data values and evaluating prediction errors; Aggregating prediction errors and visualizing them highlights regions of interesting behavior; Applicable to datasets from various domains, meaningful results are produced with minimal assumptions.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2021)
Article
Energy & Fuels
Yuan Gao, Shohei Miyata, Yasunori Akashi
Summary: With the rapid development of high-performance computing technology, data-driven models, especially deep learning models, are increasingly used for solar radiation prediction. However, the lack of interpretability in the black box models limits their application in final optimization scenarios. In this study, we proposed models based on recurrent neural network prediction model to improve the interpretability of the models through the design and improvement of the model structure, thereby increasing the credibility of the model results. The use of attention mechanism and graph neural network helped to study the interpretability in time and spatial dependencies of the prediction process. Results showed that the deep learning model, with attention, could effectively adapt to varying prediction target hours, and the graph neural network identified the most relevant variables related to solar radiation.
Article
Energy & Fuels
Yuan Gao, Shohei Miyata, Yasunori Akashi
Summary: The study focuses on enhancing the interpretability of deep learning models through the design and improvement of model structure, investigating time and spatial dependencies of the prediction process using attention mechanisms and graph neural networks.
Article
Green & Sustainable Science & Technology
Gonzalo Parrado-Hernando, Luka Herc, Antun Pfeifer, Inigo Capellan-Perez, Ilija Batas Bjelic, Neven Duic, Fernando Frechoso-Escudero, Luis Javier Miguel Gonzalez, Vladimir Z. Gjorgievski
Summary: Long-term energy planning is shifting towards finer temporal and spatial resolutions to achieve decarbonization. However, current integrated assessment models lack the ability to capture the specific dynamics of supply and demand in the transition to renewable energy sources. This article presents a method to translate an hourly-resolution energy model into a yearly-resolution model and validates it using the European Union region.
Article
Engineering, Environmental
Aliasghar Azma, Yakun Liu, Masoumeh Azma, Mohsen Saadat, Di Zhang, Jinwoo Cho, Shahabaldin Rezania
Summary: Measuring water quality parameters is important for hydrological assessments, and this study proposes two intelligent models, BBO-ANN and ASO-ANN, to predict daily dissolved oxygen (DO) using biogeography-based optimization (BBO), atom search optimization (ASO), and artificial neural network (ANN). The models are compared with benchmark techniques and validated using five-year water quality data from Rock Creek station. The results show that the models can accurately predict DO, with MAPEs of around 4% and correlations of 97.5%. The BBO-ANN and ASO-ANN models outperform similar hybrids in the literature.
JOURNAL OF WATER PROCESS ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Aydin Dogan, Engin Demir
Summary: This study introduces novel models using the structural recurrent neural network (SRNN) to capture the spatial proximity and structural properties in earthquake prediction. Experimental results in two distinct regions, Turkey and China, show that the SRNN models achieve better performance compared to baseline and state-of-the-art models. Particularly, the SRNNClass(near) model, which captures the first-order spatial neighborhood and structural classification based on fault lines, achieves the highest F-1 score.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Engineering, Multidisciplinary
Fahad Aljuaydi, Benchawan Wiwatanapataphee, Yong Hong Wu
Summary: This paper presents multivariate machine learning-based prediction models for freeway traffic flow under non-recurrent events. Five different model architectures, including MLP, CNN, LSTM, CNN-LSTM, and Autoencoder LSTM networks, are developed to predict traffic flow under road crashes and rainfall. The models' performance is evaluated using an input dataset with five features (flow rate, speed, density, road incident, and rainfall) and two standard metrics (Root Mean Square error and Mean Absolute error).
ALEXANDRIA ENGINEERING JOURNAL
(2023)
Article
Computer Science, Interdisciplinary Applications
Zhiyan Yi, Xiaoyue Cathy Liu, Nikola Markovic, Jeff Phillips
Summary: This paper introduces an innovative spatial prediction method for hourly traffic volume on a network scale, utilizing the XGBoost tree ensemble model to handle large-scale features and hourly traffic volume samples. In addition, spatial dependency among road segments is considered using graph theory and a graph-based approach in the proposed model. The results from testing on Utah's road network show high computational efficiency and significant improvement in prediction accuracy compared to benchmarked models.
COMPUTERS ENVIRONMENT AND URBAN SYSTEMS
(2021)
Article
Environmental Sciences
Chao Song, Xiaohong Chen
Summary: This paper focuses on improving precipitation prediction accuracy through the use of different decomposition methods to build prediction models, using annual precipitation in Guangzhou as a case study. The TVF-EMD-ENN model shows the best prediction performance, with secondary decomposition significantly improving accuracy.
Article
Computer Science, Artificial Intelligence
Licheng Qu, Jiao Lyu, Wei Li, Dongfang Ma, Haiwei Fan
Summary: Accurate traffic speed forecasting is critical and the proposed feature injected recurrent neural networks (FI-RNNs) show great potential in improving prediction accuracy by combining sequential time data with contextual factors. Case studies on real-world data sets demonstrate that the injection of contextual features can greatly enhance the accuracy of time series prediction, outperforming other state-of-the-art traffic prediction methods.
Article
Environmental Sciences
Guowen Huang
Summary: This study compares and evaluates the performance of hourly and daily modeling strategies in estimating daily pollution exposure, and the results consistently show that daily pollution models perform better.
ATMOSPHERIC ENVIRONMENT
(2023)
Article
Engineering, Civil
Sakib Mahmud Khan, Kakan C. Dey, Mashrur Chowdhury
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2017)
Article
Engineering, Civil
Mashrur Chowdhury, Mizanur Rahman, Anjan Rayamajhi, Sakib Mahmud Khan, Mhafuzul Islam, Zadid Khan, James Martin
TRANSPORTATION RESEARCH RECORD
(2018)
Review
Engineering, Civil
Sakib Mahmud Khan, Mashrur Chowdhury, Eric A. Morris, Lipika Deka
JOURNAL OF INFRASTRUCTURE SYSTEMS
(2019)
Article
Urban Studies
Sakib Mahmud Khan, Mashrur Chowdhury, Linh B. Ngo, Amy Apon