Geographically Aware Air Quality Prediction Through CNN-LSTM-KAN Hybrid Modeling with Climatic and Topographic Differentiation DOI Creative Commons
Yue Hu, Yongkun Ding, Wenjing Jiang

и другие.

Atmosphere, Год журнала: 2025, Номер 16(5), С. 513 - 513

Опубликована: Апрель 28, 2025

Air pollution poses a pressing global challenge, particularly in rapidly industrializing nations like China where deteriorating air quality critically endangers public health and sustainable development. To address the heterogeneous patterns of across diverse geographical climatic regions, this study proposes novel CNN-LSTM-KAN hybrid deep learning framework for high-precision Quality Index (AQI) time-series prediction. Through systematic analysis multi-city AQI datasets encompassing five representative Chinese metropolises—strategically selected to cover climate zones (subtropical temperate), gradients (coastal inland), topographical variations (plains mountains)—we established three principal methodological advancements. First, Shapiro–Wilk normality testing (p < 0.05) revealed non-Gaussian distribution characteristics observational data, providing statistical justification implementing Gaussian filtering-based noise suppression. Second, our multi-regional validation extended beyond conventional single-city approaches, demonstrating model generalizability distinct environmental contexts. Third, we innovatively integrated Kolmogorov–Arnold Networks (KANs) with attention mechanisms replace traditional fully connected layers, achieving enhanced feature weighting capacity. Comparative experiments demonstrated superior performance 23.6–59.6% reduction Root-Mean-Square Error (RMSE) relative baseline LSTM models, along consistent outperformance over CNN-LSTM hybrids. Cross-regional correlation analyses identified PM2.5/PM10 as dominant predictive factors. The developed exhibited robust generalization capabilities divisions (R2 = 0.92–0.99), establishing reliable decision-support platform regionally adaptive early-warning systems. This provides valuable insights addressing spatial heterogeneity modeling applications.

Язык: Английский

The CEEMDAN-EWT-CNN-GRU-SVM Model: A Robust Framework for Decomposing Non-Stationary Time Series, Extracting Data features, and Predicting Solar Radiation DOI Creative Commons
Sharareh Pourebrahim, Akram Seifi,

Mohammad Ehteram

и другие.

Results in Engineering, Год журнала: 2025, Номер unknown, С. 104267 - 104267

Опубликована: Фев. 1, 2025

Язык: Английский

Процитировано

1
Autonomous underwater vehicle fault diagnosis model based on a deep belief rule with attribute reliability DOI
Jia Mai, Hai Huang, Wei Fan

и другие.

Ocean Engineering, Год журнала: 2025, Номер 321, С. 120472 - 120472

Опубликована: Янв. 24, 2025

Язык: Английский

Процитировано

1
Hyperspectral estimation of chlorophyll content in grapevine based on feature selection and GA-BP DOI Creative Commons
Yafeng Li, Xingang Xu,

Wenbiao Wu

и другие.

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Март 7, 2025

Abstract Leaf chlorophyll content (LCC) is a key indicator for assessing the growth of grapes. Hyperspectral techniques have been applied to LCC research. However, quantitative prediction grape using this technique remains challenging due baseline drift, spectral peak overlap, and ambiguity in sensitive range. To address these issues, two typical crop leaf hyperspectral data were collected reveal response characteristics standardization by variables (SNV) multiple far scattering correction (MSC) preprocessing variations. The range determined Pearson’s algorithm, features are further extracted within that Extreme Gradient Boosting (XGBoost), Recursive Feature Elimination (RFE), Principal components analysis (PCA). Comparison ability Random Forest Regression (RFR) Support Vector Machine (SVR) model, Genetic Algorithm-Based Neural Network (GA-BP) on based features. A SNV-RFE-GA-BP framework predicting grapes proposed, where $$\:{R}^{2}$$ =0.835 NRMSE = 0.091. results show SNV MSC treatments improve correlation between reflectance LCC, different feature screening methods greater impact model accuracy. It was shown SNV-based processed combined with GA-BP has great potential efficient monitoring grapevine. This method provides new theory constructing analytical grapevine indicators.

Язык: Английский

Процитировано

1
Machine learning prediction of surface roughness in sustainable machining of AISI H11 tool steel DOI Creative Commons
A. Balasuadhakar,

S. Thirumalai Kumaran,

M. Uthayakumar

и другие.

Smart Materials in Manufacturing, Год журнала: 2025, Номер 3, С. 100075 - 100075

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
Geographically Aware Air Quality Prediction Through CNN-LSTM-KAN Hybrid Modeling with Climatic and Topographic Differentiation DOI Creative Commons
Yue Hu, Yongkun Ding, Wenjing Jiang

и другие.

Atmosphere, Год журнала: 2025, Номер 16(5), С. 513 - 513

Опубликована: Апрель 28, 2025

Air pollution poses a pressing global challenge, particularly in rapidly industrializing nations like China where deteriorating air quality critically endangers public health and sustainable development. To address the heterogeneous patterns of across diverse geographical climatic regions, this study proposes novel CNN-LSTM-KAN hybrid deep learning framework for high-precision Quality Index (AQI) time-series prediction. Through systematic analysis multi-city AQI datasets encompassing five representative Chinese metropolises—strategically selected to cover climate zones (subtropical temperate), gradients (coastal inland), topographical variations (plains mountains)—we established three principal methodological advancements. First, Shapiro–Wilk normality testing (p < 0.05) revealed non-Gaussian distribution characteristics observational data, providing statistical justification implementing Gaussian filtering-based noise suppression. Second, our multi-regional validation extended beyond conventional single-city approaches, demonstrating model generalizability distinct environmental contexts. Third, we innovatively integrated Kolmogorov–Arnold Networks (KANs) with attention mechanisms replace traditional fully connected layers, achieving enhanced feature weighting capacity. Comparative experiments demonstrated superior performance 23.6–59.6% reduction Root-Mean-Square Error (RMSE) relative baseline LSTM models, along consistent outperformance over CNN-LSTM hybrids. Cross-regional correlation analyses identified PM2.5/PM10 as dominant predictive factors. The developed exhibited robust generalization capabilities divisions (R2 = 0.92–0.99), establishing reliable decision-support platform regionally adaptive early-warning systems. This provides valuable insights addressing spatial heterogeneity modeling applications.

Язык: Английский

Процитировано

0