AIoT-driven multi-source sensor emission monitoring and forecasting using multi-source sensor integration with reduced noise series decomposition

Journal of Cloud Computing Advances Systems and Applications, Год журнала: 2024, Номер 13(1)

Опубликована: Март 21, 2024

Abstract The integration of multi-source sensors based AIoT (Artificial Intelligence Things) technologies into air quality measurement and forecasting is becoming increasingly critical in the fields sustainable smart environmental design, urban development, pollution control. This study focuses on enhancing prediction emission, with a special emphasis pollutants, utilizing advanced deep learning (DL) techniques. Recurrent neural networks (RNNs) long short-term memory (LSTM) have shown promise predicting trends time series data. However, challenges persist due to unpredictability data scarcity long-term historical for training. To address these challenges, this introduces AIoT-enhanced EEMD-CEEMDAN-GCN model. innovative approach involves decomposing input signal using EEMD (Ensemble Empirical Mode Decomposition) CEEMDAN (Complete Ensemble Decomposition Adaptive Noise) extract intrinsic mode functions. These functions are then processed through GCN (Graph Convolutional Network) model, enabling precise trends. model’s effectiveness validated datasets from four provinces China, demonstrating its superiority over various models (GCN, EMD-GCN) decomposition (EEMD-GCN, CEEMDAN-GCN). It achieves higher accuracy better fitting, outperforming other key metrics such as MAE (Mean Absolute Error), MSE Squared MAPE Percentage R 2 (Coefficient Determination). implementation model allows decision-makers more accurately anticipate changes quality, particularly concerning carbon emissions. facilitates effective planning mitigation measures, improvement public health, optimization resource allocation. Moreover, adeptly addresses complexities data, contributing significantly enhanced monitoring management strategies context development conservation.

Язык: Английский

---

Long-Term Retrospective Predicted Concentration of PM2.5 in Upper Northern Thailand Using Machine Learning Models

Toxics, Год журнала: 2025, Номер 13(3), С. 170 - 170

Опубликована: Фев. 27, 2025

This study aims to build, for the first time, a model that uses machine learning (ML) approach predict long-term retrospective PM2.5 concentrations in upper northern Thailand, region impacted by biomass burning and transboundary pollution. The dataset includes PM10 levels, fire hotspots, critical meteorological data from 1 January 2011 31 December 2020. ML techniques, namely multi-layer perceptron neural network (MLP), support vector (SVM), multiple linear regression (MLR), decision tree (DT), random forests (RF), were used construct prediction models. best was selected considering root mean square error (RMSE), (MPE), relative (RPE) (the lower, better), coefficient of determination (R2) bigger, better). Our found model-based RF technique using PM10, CO2, O3, air pressure, rainfall, humidity, temperature, wind direction, speed performs when predicting concentration with an RMSE 6.82 µg/m3, MPE 4.33 RPE 22.50%, R2 0.93. this research could further studies effects on human health related issues.

Язык: Английский

---

Improving the quantification of fine particulates (PM2.5) concentrations in Malaysia using simplified and computationally efficient models

Journal of Cleaner Production, Год журнала: 2024, Номер 448, С. 141559 - 141559

Опубликована: Март 7, 2024

Язык: Английский

---

PD-LL-Transformer: An Hourly PM2.5 Forecasting Method over the Yangtze River Delta Urban Agglomeration, China

Remote Sensing, Год журнала: 2024, Номер 16(11), С. 1915 - 1915

Опубликована: Май 27, 2024

As the urgency of PM2.5 prediction becomes increasingly ingrained in public awareness, deep-learning methods have been widely used forecasting concentration trends and other atmospheric pollutants. Traditional time-series models, like long short-term memory (LSTM) temporal convolutional network (TCN), were found to be efficient pollutant estimation, but either model accuracy was not high enough or models encountered certain challenges due their own structure some specific application scenarios. This study proposed a high-accuracy, hourly model, poly-dimensional local-LSTM Transformer, namely PD-LL-Transformer, by methods, based on air data meteorological data, aerosol optical depth (AOD) retrieved from Himawari-8 satellite. research Yangtze River Delta Urban Agglomeration (YRDUA), China for 2020–2022. The PD-LL-Transformer had three parts: embedding layer, which integrated advantages allocating multi-variate features more refined manner combined superiority different processing methods; block, LSTM TCN; Transformer encoder block. Over test set (the whole year 2022), model’s R2 0.8929, mean absolute error (MAE) 4.4523 µg/m3, root squared (RMSE) 7.2683 showing great prediction. surpassed existing upon same tasks similar datasets, with help tool better performance applicability could established.

Язык: Английский

---

Using Random Forest to improve EMEP4PL model estimates of daily PM2.5 in Poland

Atmospheric Environment, Год журнала: 2024, Номер 332, С. 120615 - 120615

Опубликована: Май 31, 2024

Long-term exposure to poor air quality is responsible for many diseases and increased mortality worldwide. European Environmental Agency reports that Poland one of the most polluted countries in Europe due high emissions associated with large coal wood consumption specific weather conditions. Exceedances WHO-recommended PM2.5 thresholds are still common Poland, so further action needed protect health population. Atmospheric chemical transport models (CTMs) provide information on public used regulate pollutant emissions. However, uncertainties CTMs, related e.g. physical/chemical processes input data often lead underestimation concentrations, especially PM2.5, limits applicability CTMs impact studies. A hybrid approach combining EMEP4PL model Random Forest (RF) machine learning algorithm was applied address limitations CTM reduce its underestimation. We EMEP4PL-modelled concentrations period 2016-2019 as a predictor measured daily from 71 monitoring stations dependent variable three RF scenarios, which differed terms selected predictors. The different additional variables area revealed, including population emission data, dominant type land use, Weather Research Forecast (WRF) meteorological parameters, temporal patterns across years. were evaluated random 5-fold spatial leave-one-station-out cross-validations (LOSOCV), well an independent test set. Our final achieved set R2 0.71, compared 0.38 EMEP4PL, along reduction negative bias (0.25 μg m-3 RF, -11 EMEP4PL) improved ability detect severe episodes. Enhanced coefficients determination observed all seasons at sites included study, both types cross-validation estimated contribution each group separately discovered, impactful predictors calculated based averages outcome (such day year, week number, etc.) modelled factors temperature, planetary boundary layer height, wind speed, atmospheric pressure. developed provides basis spatiotemporal estimates forecasting region, important step toward better understanding pollution local well-being.

Язык: Английский

---

Understanding the Disparities of PM2.5 Air Pollution in Urban Areas via Deep Support Vector Regression

Environmental Science & Technology, Год журнала: 2024, Номер 58(19), С. 8404 - 8416

Опубликована: Май 3, 2024

In densely populated urban areas, PM2.5 has a direct impact on the health and quality of residents' life. Thus, understanding disparities is crucial for ensuring sustainability public health. Traditional prediction models often overlook spillover effects within areas complexity data, leading to inaccurate spatial predictions PM2.5. We propose Deep Support Vector Regression (DSVR) that as graph, with grid center points nodes connections between grids edges. Nature human activity features each are initialized representation node. Based DSVR uses random diffusion-based deep learning quantify It leverages walk uncover more extensive relationships nodes, thereby capturing both local nonlocal And then it engages in predictive using feature vectors encapsulate effects, enhancing across different regions. By applying our proposed model northern region New York performance analysis, we found consistently outperforms other models. During periods surges, R-square reaches high 0.729, outperforming non-spillover by 2.5 5.7 times traditional metric 2.2 4.6 times. Therefore, holds significant importance air pollution taking first steps toward new method considers nonlinear data prediction.

Язык: Английский

---

A comprehensive evaluation model for forest fires based on MCDA and machine learning: A case study of Zhenjiang City, China

Environment Development and Sustainability, Год журнала: 2024, Номер unknown

Опубликована: Июнь 11, 2024

Язык: Английский

---

Comparative analysis of machine learning algorithms for air quality index prediction

Machine learning for computational science and engineering, Год журнала: 2025, Номер 1(1)

Опубликована: Фев. 25, 2025

Язык: Английский

---

Bayesian modelling for the integration of spatially misaligned health and environmental data

Stochastic Environmental Research and Risk Assessment, Год журнала: 2025, Номер unknown

Опубликована: Март 8, 2025

Язык: Английский

---

Grazing intensity estimation in temperate typical grasslands of Inner Mongolia using machine learning models

Ecological Indicators, Год журнала: 2025, Номер 172, С. 113318 - 113318

Опубликована: Март 1, 2025

Язык: Английский

---