Air Pollution Forecasting Using a Deep Learning Model Based on 1D ConvNets and Bidirectional GRU

International Journal for Research in Applied Science and Engineering Technology, Год журнала: 2024, Номер 12(6), С. 1808 - 1820

Опубликована: Июнь 28, 2024

Abstract: This study presents a methodology for air pollution forecasting, aiming to improve accuracy through incorporating models deep learning. Our goal is create reliable technique forecasting concentrations. Central Pollution Control Board (CPCB) data undergoes exploratory analysis. Analysis (EDA) and pre-processing before being split into training testing sets. Two sequential models, Sequential-1 Sequential-2, are compared, with Sequential-2 Conv1D layers alongside GRU enhanced spatial-temporal modeling. Findings reveal that consistently outperforms Sequential-1, exhibiting lower loss, mean squared error (MSE), validation MSE metrics. indicates Sequential-2's superior predictive performance generalization capability, attributed because of how well it can grasp spatial dependencies. In sum, the proves learning methods work predicting levels, offering promising avenues accurately pollutant concentrations informing mitigation strategies healthier environment

Язык: Английский

---

Prediction of air quality levels to support sustainable development goal – 11 using multiple deep learning classifiers

Smart and Sustainable Built Environment, Год журнала: 2025, Номер unknown

Опубликована: Март 24, 2025

Purpose Sustainable Development Goal (SDG) 11 emphasizes the importance of monitoring air quality to develop cities that are resilient, safe and sustainable on a global scale. Particulate matter pollutants such as PM2.5 PM10 have detrimental impact both human health environment. Traditional methods for assessing often face challenges related scalability accuracy. This paper aims introduce an automated system designed predict levels (AQLs). These categorized good, moderate, unhealthy hazardous, based index. Design/methodology/approach uses dataset 8.1 million records from various US cities. The data undergoes preprocessing remove inconsistencies ensure uniformity. Scaling techniques applied standardize values across dataset. Augmentation methods, including K Nearest Neighbour, z -score normalization Synthetic Minority Oversampling Technique (SMOTE), employed balance enhance Later, used train eight deep learning models, standard, bidirectional stacked architectures. Additionally, two hybrid models also developed by combining features different Findings validation results demonstrate system’s exceptional performance. Bidirectional GRU model achieves highest accuracy 99.98%. Similarly, RNN + impressive 99.92%. Furthermore, Stacked Gated Recurrent Unit stands out, achieving perfect scores 100% precision, recall F1 score. Originality/value assessment approaches rely heavily basic statistical limited scope their datasets. In contrast, this study presents innovative methodology employs advanced By incorporating sophisticated techniques, proposed significantly enhances detection classification AQLs, setting new benchmark development objectives.

Язык: Английский

---

African buffalo optimization with deep learning-based intrusion detection in cyber-physical systems

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

Опубликована: Март 25, 2025

Cyber-physical system (CPS) incorporates several computing resources, networking units, interconnected physical processes, and monitoring the development application of system. Interconnection between cyber worlds initiates attacks on security problems, particularly with enhancing complications transmission networks. Despite efforts to combat these analyzing detecting cyber-physical from complex CPS is challenging. Machine learning (ML)-researcher workers implemented based techniques examine systems. A competent network intrusion detection (IDS) essential avoid attacks. Generally, IDS uses ML classify However, features used for classification are not frequently appropriate or adequate. Moreover, number intrusions much lower than that non-intrusions. This research presents an African Buffalo Optimizer Algorithm a Deep Learning Intrusion Detection (ABOADL-IDS) model in environment. The main intention ABOADL-IDS utilize FS optimal DL approach recognition identification procedure. Initially, performs data normalization process. Furthermore, utilizes ABO technique feature selection. stacked deep belief (SDBN) employed identification. To improve SDBN solution, seagull optimization (SGO) hyperparameter assessment accomplished under NSLKDD2015 CICIDS2015 datasets. performance validation illustrated superior accuracy value 99.28% over existing models concerning various measures.

Язык: Английский

---

Active learning regression quality prediction model and grinding mechanism for ceramic bearing grinding processing

PLoS ONE, Год журнала: 2025, Номер 20(4), С. e0320494 - e0320494

Опубликована: Апрель 7, 2025

The study aims to explore quality prediction in ceramic bearing grinding processing, with particular focus on the effect of parameters surface roughness. uses active learning regression model for construction and optimization, empirical analysis under different conditions. At same time, various deep models are utilized conduct experiments processing. experimental setup covers a variety parameters, including wheel linear speed, depth feed rate, ensure accuracy reliability According results, when increases 21 μm, average training loss further decreases 0.03622, roughness Ra value significantly 0.1624 μm. In addition, experiment also found that increasing velocity moderately adjusting can improve machining quality. For example, is 45 m/s 0.015 mm, drops 0.1876 results not only provide theoretical support processing bearings, but basis optimization actual production, which has an important industrial application value.

Язык: Английский

---

Red Kite Optimization Algorithm With Average Ensemble Model for Intrusion Detection for Secure IoT

IEEE Access, Год журнала: 2023, Номер 11, С. 131749 - 131758

Опубликована: Янв. 1, 2023

The Internet of Things (IoT) based Wireless Sensor Networks (WSNs) contain interconnected autonomous sensor nodes (SN), which wirelessly communicate with each other and the wider internet structure. Intrusion detection to secure IoT-based WSNs is critical for identifying responding great security attacks threats that can cooperate integrity, availability, privacy network its data. Machine learning (ML) algorithms are deployed detecting difficult patterns subtle anomalies in IoT Artificial intelligence (AI) driven methods learned adapted from novel data improving accuracy over time. In this article, we introduce a Red Kite Optimization Algorithm an Average Ensemble Model Detection (RKOA-AEID) technique Secure WSN. purpose RKOA-AEID methodology accomplish solutions IoT-assisted WSNs. To this, performs pre-processing scale input using min-max normalization. addition, RKOA-based feature selection approach elect optimum set features. For intrusion detection, average ensemble model used. Finally, Lévy-fight chaotic whale optimization (LCWOA) be executed hyperparameter chosen models. performance evaluation algorithm tested on benchmark WSN-DS dataset. extensive experimental outcomes stated higher outcome approaches improved 98.94%.

Язык: Английский

---

Marine Ecological Environment Monitoring and Management System Based on Sensor Technology

Опубликована: Фев. 23, 2024

The Ocean is one of the most important ecosystems on Earth. However, continuous expansion and intensification human activities has caused serious threats to marine ecological environment. In addition, increasing demand for development resources also made environment less stable. order achieve sustainable ocean, it necessary conduct real-time dynamic detection control ocean Current monitoring technical weaknesses complex. protect improve environment, in-depth research sensors based existing technologies performance management systems. This article discusses network nodes involved in sensor technology, analyzes composition several sensors, then designs a system emphasizes importance proposes transmission integration data. Finally, this debugs simulates compare difference between predicted value actual value. experimental results show that, overall, there certain error ocean-related data collected by applying value, but proportion (less than 2%) not obvious.

Язык: Английский

---

Hybrid model for air quality prediction based on LSTM with random search and Bayesian optimization techniques

Earth Science Informatics, Год журнала: 2024, Номер 18(1)

Опубликована: Дек. 12, 2024

Язык: Английский

---

Harnessing Deep Learning and Snow Cover Data for Enhanced Runoff Prediction in Snow-Dominated Watersheds

Atmosphere, Год журнала: 2024, Номер 15(12), С. 1407 - 1407

Опубликована: Ноя. 22, 2024

Predicting streamflow is essential for managing water resources, especially in basins and watersheds where snowmelt plays a major role river discharge. This study evaluates the advanced deep learning models accurate monthly peak forecasting Gilgit River Basin. The utilized were LSTM, BiLSTM, GRU, CNN, their hybrid combinations (CNN-LSTM, CNN-BiLSTM, CNN-GRU, CNN-BiGRU). Our research measured model’s accuracy through root mean square error (RMSE), absolute (MAE), Nash–Sutcliffe efficiency (NSE), coefficient of determination (R2). findings indicated that models, CNN-BiGRU achieved much better performance than traditional like LSTM GRU. For instance, lowest RMSE (71.6 training 95.7 testing) highest R2 (0.962 0.929 testing). A novel aspect this was integration MODIS-derived snow-covered area (SCA) data, which enhanced model substantially. When SCA data included, CNN-BiLSTM improved from 83.6 to 71.6 during 108.6 testing. In prediction, outperformed other with (108.4), followed by (144.1). study’s results reinforce notion combining CNN’s spatial feature extraction capabilities temporal dependencies captured or GRU significantly enhances accuracy. demonstrated improvements prediction accuracy, extreme events, highlight potential these support more informed decision-making flood risk management allocation.

Язык: Английский

---

Human Action Recognition and Analysis Methods Based on OpenPose and Deep Learning

Опубликована: Фев. 23, 2024

Currently, commonly used human action recognition (HAR) methods include two categories: based on manual features and machine learning. However, these traditional often rely handcrafted features, which require extensive domain knowledge may not capture all the intricacies of actions. To address this limitation, article proposes a novel approach that combines key technologies in OpenPose, state-of-the-art pose estimation algorithm, with deep learning techniques. By leveraging rich spatial temporal information provided by proposed method can fine-grained details actions higher accuracy efficiency. The component further enhances performance automatically discriminative from input data. Experimental results show application increase HAR to maximum 95.6%. Therefore, it be determined OpenPose has high accuracy, precision recall rates HAR.

Язык: Английский

---

Human Resource Management Decision Support System Based on Multi Agent

Опубликована: Март 15, 2024

Due to the rapid development of economy and increasing level management, human resource management is increasingly valued by enterprise managers has become a very important essential part management. However, many traditional methods are still based on experience subjective judgment, lacking statistical or data analysis support, which can easily lead inaccurate decision-making waste resources. This article aims design decision support system that alleviate secondary problems. The second paragraph this introduces current research third structure system, fourth tests designed in achieves good results. Further needed address issue

Язык: Английский

---