Enhanced Sequence-to-Sequence Attention-Based PM2.5 Concentration Forecasting Using Spatiotemporal Data DOI Creative Commons
Baekcheon Kim, Eun Kyeong Kim, Seunghwan Jung

et al.

Atmosphere, Journal Year: 2024, Volume and Issue: 15(12), P. 1469 - 1469

Published: Dec. 9, 2024

Severe air pollution problems continue to increase because of accelerated industrialization and urbanization. Specifically, fine particulate matter (PM2.5) causes respiratory cardiovascular diseases, according the World Health Organization (WHO), millions premature deaths significant health burdens annually. Therefore, PM2.5 concentration forecasting is essential. This study proposed a method forecast concentrations one hour after using Sequence-to-Sequence Attention (Seq2Seq-attention). The selects neighboring stations minimum redundancy maximum relevance (mRMR) integrates their data convolutional neural network (CNN). attention score Seq2Seq are used on integrated hour. performance validated through two case studies. first comparison evaluated conventional against scores. second results with without considering stations. showed that improved index (Root Mean Square Error (RMSE): 3.48%p, Absolute (MAE): 8.60%p, R2: 0.49%p, relative Root (rRMSE): 3.64%p, Percent Bias (PBIAS): 59.29%p). stations’ can be more effective in than standalone station (RMSE: 5.49%p, MAE: 0.51%p, 0.67%p, rRMSE: 5.44%p, PBIAS: 46.56%p). confirmed effectively

Language: Английский

Development of a novel modeling framework based on weighted kernel extreme learning machine and ridge regression for streamflow forecasting DOI Creative Commons
Arvin Samadi-Koucheksaraee, Xuefeng Chu

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Dec. 28, 2024

A precise streamflow forecast is crucial in hydrology for flood alerts, water quantity and quality management, disaster preparedness. Machine learning (ML) techniques are commonly employed hydrological prediction; however, they still face certain drawbacks, such as the need to optimize appropriate predictors, ability of models generalize across different time horizons, analysis high-dimensional series. This research aims address these specific drawbacks by developing a novel framework forecasting. Specifically, hybrid ML model, WKELM-R, developed predict based on daily discharge precipitation. The model combines ridge regression (RR), locally weighted linear (LWLR), kernel extreme machine (KELM) enhance multi-step-ahead predictions accounting both nonlinear characteristics. In data preprocessing, this study applies multivariate variational mode decomposition (MVMD) handle non-stationarity complexity, Boruta-XGBoost feature selection select optimal inputs decrease dimension, gradient-based optimizer (GBO) adjustment parameters overcome predictors. To demonstrate real-world conditions WKELM-R was applied watershed North Dakota, USA three horizons. results were compared with those from existing standalone multi-criteria decision-making (MCDM), demonstrating efficacy unique capabilities new forecasting (for testing level at t + 3: R = 0.992, RMSE 0.426, NSE 0.983; 7: 0.997, 0.249, 0.994; 14: 0.996, 0.304, 0.991).

Language: Английский

Citations

2

Enhanced Sequence-to-Sequence Attention-Based PM2.5 Concentration Forecasting Using Spatiotemporal Data DOI Creative Commons
Baekcheon Kim, Eun Kyeong Kim, Seunghwan Jung

et al.

Atmosphere, Journal Year: 2024, Volume and Issue: 15(12), P. 1469 - 1469

Published: Dec. 9, 2024

Severe air pollution problems continue to increase because of accelerated industrialization and urbanization. Specifically, fine particulate matter (PM2.5) causes respiratory cardiovascular diseases, according the World Health Organization (WHO), millions premature deaths significant health burdens annually. Therefore, PM2.5 concentration forecasting is essential. This study proposed a method forecast concentrations one hour after using Sequence-to-Sequence Attention (Seq2Seq-attention). The selects neighboring stations minimum redundancy maximum relevance (mRMR) integrates their data convolutional neural network (CNN). attention score Seq2Seq are used on integrated hour. performance validated through two case studies. first comparison evaluated conventional against scores. second results with without considering stations. showed that improved index (Root Mean Square Error (RMSE): 3.48%p, Absolute (MAE): 8.60%p, R2: 0.49%p, relative Root (rRMSE): 3.64%p, Percent Bias (PBIAS): 59.29%p). stations’ can be more effective in than standalone station (RMSE: 5.49%p, MAE: 0.51%p, 0.67%p, rRMSE: 5.44%p, PBIAS: 46.56%p). confirmed effectively

Language: Английский

Citations

0