Deep-learning architecture for PM2.5 concentration prediction: A review

Accurately predicting the concentration of fine particulate matter (PM2.5) is crucial for evaluating air pollution levels and public exposure. Recent advancements have seen a significant rise in using deep learning (DL) models for forecasting PM2.5 concentrations. Nonetheless, there is a lack of unified and standardized frameworks for assessing the performance of DL-based PM2.5 prediction models. Here we extensively reviewed those DL-based hybrid models for forecasting PM2.5 levels according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. We examined the similarities and differences among various DL models in predicting PM2.5 by comparing their complexity and effectiveness. We categorized PM2.5 DL methodologies into seven types based on performance and application conditions, including four types of DL-based models and three types of hybrid learning models. Our research indicates that established deep learning architectures are commonly used and respected for their efficiency. However, many of these models often fall short in terms of innovation and interpretability. Conversely, models hybrid with traditional approaches, like deterministic and statistical models, exhibit high interpretability but compromise on accuracy and speed. Besides, hybrid DL models, representing the pinnacle of innovation among the studied models, encounter issues with interpretability. We introduce a novel three-dimensional evaluation framework, i.e., Dataset-Method-Experiment Standard (DMES) to unify and standardize the evaluation for PM2.5 predictions using DL models. This review provides a framework for future evaluations of DL-based models, which could inspire researchers to standardize DL model usage in PM2.5 prediction and improve the quality of related studies.

Spatio-temporal fusion of meteorological factors for multi-site PM2.5 prediction: A deep learning and time-variant graph approach

In the field of environmental science, traditional methods for predicting PM2.5 concentrations primarily focus on singular temporal or spatial dimensions. This approach presents certain limitations when it comes to deeply mining the joint influence of multiple monitoring sites and their inherent connections with meteorological factors. To address this issue, we introduce an innovative deep-learning-based multi-graph model using Beijing as the study case. This model consists of two key modules: firstly, the 'Meteorological Factor Spatio-Temporal Feature Extraction Module'. This module deeply integrates spatio-temporal features of hourly meteorological data by employing Graph Convolutional Networks (GCN) and Long Short-Term Memory (LSTM) for spatial and temporal encoding respectively. Subsequently, through an attention mechanism, it retrieves a feature tensor associated with air pollutants. Secondly, these features are amalgamated with PM2.5 concentration values, allowing the 'PM2.5 Concentration Prediction Module' to predict with enhanced accuracy the joint influence across multiple monitoring sites. Our model exhibits significant advantages over traditional methods in processing the joint impact of multiple sites and their associated meteorological factors. By providing new perspectives and tools for the in-depth understanding of urban air pollutant distribution and optimization of air quality management, this model propels us towards a more comprehensive approach in tackling air pollution issues.

Prediction of PM2.5 time series by seasonal trend decomposition-based dendritic neuron model

The rapid industrial development in the human society has brought about the air pollution, which seriously affects human health. PM2.5 concentration is one of the main factors causing the air pollution. To accurately predict PM2.5 microns, we propose a dendritic neuron model (DNM) trained by an improved state-of-matter heuristic algorithm (DSMS) based on STL-LOESS, namely DS-DNM. Firstly, DS-DNM adopts STL-LOESS for the data preprocessing to obtain three characteristic quantities from original data: seasonal, trend, and residual components. Then, DNM trained by DSMS predicts the residual values. Finally, three sets of feature quantities are summed to obtain the predicted values. In the performance test experiments, five real-world PM2.5 concentration data are used to test DS-DNM. On the other hand, four training algorithms and seven prediction models were selected for comparison to verify the rationality of the training algorithms and the accuracy of the prediction models, respectively. The experimental results show that DS-DNM has the more competitive performance in PM2.5 concentration prediction problem.

Scientometric and multidimensional contents analysis of PM2.5 concentration prediction

The foundation for the environmental department to take suitable measures and make a significant contribution towards improving air quality is the precise and dependable prediction of PM_2.5 concentration. It is essential to review the development process and hotspots of PM_2.5 concentration prediction studies over the past 20 years (2000-2021) comprehensively and quantitatively. This study used detailed bibliometric methods and CiteSpace software to visually analyze the PM_2.5 pollution level. The outcomes found that the prediction research phases of PM_2.5 can be broadly divided into three phases and enter the rapid growth phase after 2017. Five categories of keywords are clustered, and the forecasting data and forecasting methods are typical cluster representatives. Then, the construction and processing of PM_2.5 concentration prediction datasets, the prediction methods and technical processes, and the determination of the prediction spatial-temporal scales are the main content of the analysis. In the future, it is necessary to concentrate on multi-source data fusion for PM_2.5 concentration prediction at multiple spatial-temporal scales and focus on technology integration and innovative applications in forecasting models, especially the optimal use of deep machine learning methods to improve prediction accuracy and practical application conversion.

A novel hybrid prediction model for PM2.5 concentration based on decomposition ensemble and error correction

PM_2.5 concentration is an important index to measure the degree of air pollution. It is necessary to establish an accurate PM_2.5 concentration prediction system for urban air monitoring and control. Due to the nonlinear characteristics of PM_2.5 concentration, it is difficult to predict it directly. Therefore, a novel hybrid model for PM_2.5 concentration based on improved variational mode decomposition (IVMD), outlier-robust extreme learning machine (ORELM) optimized by hybrid cuckoo search (CS), and chimp optimization algorithm (ChOA), error correction (EC) is proposed named IVMD-ChOACS-ORELM-EC. First of all, an improved VMD based on energy loss coefficient, named IVMD, is proposed. IVMD decomposes the original data to obtain K IMF components. Then, a hybrid optimization algorithm based on ChOA improved by CS is proposed, named ChOACS. The hybrid optimization algorithm is used to optimize ORELM. On this basis, the prediction model ChOACS-ORELM is proposed, and the K IMF components are predicted by ChOACS-ORELM. Finally, the EC model based on decomposition ensemble is established to further improve the prediction accuracy. The PM_2.5 concentration data collected at hourly intervals in Beijing, Shanghai, Shenyang, and Qingdao in China are used as experimental data. The experimental results show that the correlation coefficients between the prediction results and the actual values of the four cities are 0.9999, and the prediction performance of the proposed model is better than that of all comparison models.

A novel PM2.5 concentrations probability density prediction model combines the least absolute shrinkage and selection operator with quantile regression

PM2.5 has a significant negative impact on human health and atmospheric quality, and accurate prediction of its concentration is necessary. When using common point prediction models for PM2.5 concentration prediction, the influence of various uncertainties on PM2.5 concentrations makes the prediction results suffer from poor accuracy. To address this issue, this paper proposes the quantile regression neural network (QRNN) model based on the least absolute shrinkage and selection operator (LASSO), combined with kernel density estimation (KDE) for probabilistic density prediction of PM2.5 concentrations. The model uses LASSO regression to select the influential factors, and then the quartiles of daily PM2.5 concentrations calculated by the QRNN model are imported into the KDE model to obtain the probability density predictions of PM2.5 concentrations. In the paper, empirical analyses are carried out with the cities of Beijing and Jinan in China as well as six other datasets, and the prediction performance of the model is assessed by using evaluation criteria in both point prediction and interval prediction. The simulation reveals that the predictive performance of the LASSO-QRNN-KDE model is well, and the model is not only effective in filtering high-dimensional data, but also has a higher accuracy compared to common research models. In addition, the model is able to describe the uncertainty of PM2.5 concentration fluctuations and carry more information on the variation of PM2.5 concentrations, which can provide a novel and excellent PM2.5 concentration prediction tool for relevant policy makers.

A new hybrid optimization prediction model for PM2.5 concentration considering other air pollutants and meteorological conditions

With the development of economy, the problem of air pollution has become increasingly serious. As an important detection index of air pollutants, how to accurately and effectively predict PM2.5 concentration is a significant issue related to human health and development. In this paper, a new hybrid optimization prediction model for PM2.5 concentration based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), variational mode decomposition optimized by COOT optimization algorithm (COOT-VMD), and least square support vector machine (LSSVM) optimized by the JAYA optimization algorithm (JAYA-LSSVM), named CEEMDAN-COOT-VMD-JAYA-LSSVM, is proposed. To avoid artificially setting the limits of the decomposition layer and penalty factor of VMD parameters, an improved VMD by COOT optimization algorithm, named COOT-VMD, is proposed. First, the original sequence of PM2.5 concentration is decomposed by CEEMDAN. Second, the high complexity component with low prediction accuracy after once decomposition is decomposed by COOT-VMD. Third, a prediction model of optimized LSSVM by JAYA optimization algorithm, named JAYA-LSSVM, is proposed. JAYA-LSSVM is used to predict all components considering other air pollutants such as PM10, SO2, NO2, CO, and O3 and meteorological conditions such as wind speed, temperature, sunlight, relative humidity and average air pressure. Finally, all predicted values are reconstructed to obtain the final prediction results. PM2.5 concentration from April 1, 2016 to March 29, 2021 in Xi'an and Shenyang is used as the experimental data to verify the proposed model. The results of experiment in Xi'an show that the RMSE, MAE, MAPE and R² are 2.843, 1.8344, 2.94%, and 0.99525 respectively. The results of experiment in Shenyang show that the RMSE, MAE, MAPE and R² are 2.2714, 1.673, 3.13%, and 0.99573 respectively. Compared to other single and hybrid models, the proposed model can accurately predict PM2.5 concentration. Diebold Mariano test results display the proposed prediction model is superior to all comparison models at 99% confidence level.

A new hybrid prediction model of PM2.5 concentration based on secondary decomposition and optimized extreme learning machine

As air pollution worsens, the prediction of PM_2.5 concentration becomes increasingly important for public health. This paper proposes a new hybrid prediction model of PM_2.5 concentration based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), amplitude-aware permutation entropy (AAPE), variational mode decomposition improved by marine predators algorithm (MPA-VMD), and extreme learning machine optimized by chimp optimization algorithm (ChOA-ELM), named CEEMDAN-AAPE-MPA-VMD-ChOA-ELM. Firstly, CEEMDAN is used to decompose the original data, and AAPE is used to quantify the complexity of all IMF components. Secondly, MPA-VMD is used to decompose the IMF component with the maximum AAPE. Lastly, ChOA-ELM is used to predict all IMF components, and all prediction results are reconstructed to obtain the final prediction results. The proposed model combines the advantages of secondary decomposition technique, feature analysis, and optimization algorithm, which can predict PM_2.5 concentration accurately. PM_2.5 concentrations at hourly intervals collected from March 1, 2021, to March 31, 2021, in Shanghai and Shenyang, China, are used for experimental study and DM test. The experimental results in Shanghai show that the RMSE, MAE, MAPE, and R² of the proposed model are 1.0676, 0.7685, 0.0181, and 0.9980 respectively, which is better than all comparison models at 90% confidence level. In Shenyang, the RMSE, MAE, MAPE, and R² of the proposed model are 1.4399, 1.1258, 0.0389, and 0.9976, respectively, which is better than all comparison models at 95% confidence level.

PM2.5 concentration forecasting at surface monitoring sites using GRU neural network based on empirical mode decomposition

The main component of haze is the particulate matter (PM) 2.5. How to explore the laws of PM2.5 concentration changes is the main content of air quality prediction. Combining the characteristics of temporality and non-linearity in PM2.5 concentration series, more and more deep learning methods are currently applied to PM2.5 predictions, but most of them ignore the non-stationarity of time series, which leads to a lower accuracy of model prediction. To address this issue, an integration method of gated recurrent unit neural network based on empirical mode decomposition (EMD-GRU) for predicting PM2.5 concentration was proposed in this paper. This method uses empirical mode decomposition (EMD) to decompose the PM2.5 concentration sequence first and then fed the multiple stationary sub-sequences obtained after the decomposition and the meteorological features into the constructed GRU neural network successively for training and predicting. Finally, the sub-sequences of the prediction output are added to obtain the prediction results of PM2.5 concentration. The forecast result of the case in this paper show that the EMD-GRU model reduces the RMSE by 44%, MAE by 40.82%, and SMAPE by 11.63% compared to the single GRU model.

Spatio-temporal variation and daily prediction of PM2.5 concentration in world-class urban agglomerations of China

The contradiction between the development of urban agglomerations and ecological protection has long been a challenging issue. China has experienced an astonishing expansion of its urban scale in the past 40 years, and nearly 783 million of the nation's people now live in cities. Beijing-Tianjin-Hebei, the Yangtze River Delta and the Pearl River Delta have been prioritized to become world-class clusters by 2020. The health effects of air pollution in these three urban agglomerations are becoming increasingly formidable. Given these conditions, using the daily mean PM_2.5 concentration in 40 cities from January 2014 to December 2016, this research explored the spatial-temporal characteristics of PM_2.5 concentrations in these three urban agglomerations. The annual mean PM_2.5 concentrations in Beijing-Tianjin-Hebei, the Yangtze River Delta and the Pearl River Delta are 35.39 µg/m³, 53.72 µg/m³ and 78.54 µg/m³, respectively. Compared with the other two urban agglomerations, abundant rainfall causes the Pearl River Delta to have the lowest PM_2.5 level. Furthermore, a general regression neural network (GRNN) method is developed to predict the PM_2.5 concentration in these clusters on the second day, with inputs including the average, maximum and minimum temperature; average, maximum and minimum atmosphere; total rainfall; average humidity; average and maximum wind speed; and the PM_2.5 concentration measured 1 day ahead. The results indicate that the GRNN method can precisely predict the concentration level in these clusters, and it is especially useful for the Pearl River Delta, as the underlying influence mechanism is more specified in this cluster than in the others. Importantly, this 1-day-ahead forecasting of PM_2.5 concentrations can raise awareness among the public to improve their precautionary behaviours and help urban planners to provide corresponding support.