Attention-based parallel networks (APNet) for PM2.5 spatiotemporal prediction

PM2.5 concentration prediction algorithm integrating traffic congestion index

In this study, a strategy is proposed to use the congestion index as a new input feature. This approach can reveal more deeply the complex effects of traffic conditions on variations in particulate matter (PM2.5) concentrations. To assess the effectiveness of this strategy, we conducted an ablation experiment on the congestion index and implemented a multi-scale input model. Compared with conventional models, the strategy reduces the root mean square error (RMSE) of all benchmark models by > 6.07 % on average, and the best-performing model reduces it by 12.06 %, demonstrating excellent performance improvement. In addition, even with high traffic emissions, the RMSE during peak hours is still below 9.83 µg/m3, which proves the effectiveness of the strategy by effectively addressing pollution hotspots. This study provides new ideas for improving urban environmental quality and public health and anticipates inspiring further research in this domain.

Novel hybrid data-driven modeling based on feature space reconstruction and multihead self-attention gated recurrent unit: applied to PM2.5 concentrations prediction

In response to the problem of neglecting the periodic and global characteristics of sequence data when predicting PM2.5 concentrations via machine learning models, a PM2.5 concentrations prediction model based on feature space reconstruction and multihead self-attention gated recurrent unit (FSR-MSAGRU) is proposed in this study. First, the raw sequence data are subjected to frequency spectrum analysis to determine the period value of the PM2.5 sequence data. Subsequently, the seasonal trend decomposition procedure based on loess (STL) is employed to capture the periodicity and trend information in the PM2.5 sequence data. Then, the feature space of the PM2.5 sequence data is reconstructed using the raw PM2.5 sequence data, decomposed seasonal components, trend components, and residual components. Finally, the reconstructed feature data are input into multihead self-attention gated recurrent unit (MSAGRU) with the ability to capture global feature information to predict PM2.5 concentrations. Favorable prediction results were attained by the proposed FSR-MSAGRU model across 6 distinct experimental datasets, with a PCC exceeding 0.98 and a decrease in the prediction accuracy metric SMAPE of at least 68% compared to that of the GRU model. Comparative experimental results with 13 reference models demonstrate that the proposed model exhibits better prediction performances and stronger generalization abilities.

PM2.5 concentration prediction using a whale optimization algorithm based hybrid deep learning model in Beijing, China

PM2.5 is a significant global atmospheric pollutant impacting visibility, climate, and public health. Accurate prediction of PM2.5 concentrations is critical for assessing air pollution risks and providing early warnings for effective management. This study proposes a novel hybrid machine learning model that combines the whale optimization algorithm (WOA) with a convolutional neural network (CNN), long short-term memory (LSTM), and an attention mechanism (AM) to predict daily PM2.5 concentrations. Tested with meteorological and air pollution daily data from 2014 to 2018, the WOA-CNN-LSTM-AM model demonstrates substantial improvements. It achieves MAE, RMSE, MBE, and R2 values of 14.29, 21.96, -0.23, and 0.93, respectively, showing a reduction in prediction errors by 39% compared to CNN and 34% compared to LSTM models. In the medium-term forecast, the accuracy of the hybrid model is 30%-54% over WOA-CNN-LSTM and 26%-39% over CNN-LSTM-AM. The R2 value decreases by 2.5% from the 1-day to 5-day forecast, maintaining high accuracy. SHAP analysis reveals that NO2 and CO are the primary drivers for PM2.5 predictions. This study provides a reliable tool for short and medium-term PM2.5 prediction and air pollution control.

Enhancing PM2.5 prediction by mitigating annual data drift using wrapped loss and neural networks

In many deep learning tasks, it is assumed that the data used in the training process is sampled from the same distribution. However, this may not be accurate for data collected from different contexts or during different periods. For instance, the temperatures in a city can vary from year to year due to various unclear reasons. In this paper, we utilized three distinct statistical techniques to analyze annual data drifting at various stations. These techniques calculate the P values for each station by comparing data from five years (2014-2018) to identify data drifting phenomena. To find out the data drifting scenario those statistical techniques and calculate the P value from those techniques to measure the data drifting in specific locations. From those statistical techniques, the highest drifting stations can be identified from the previous year's datasets To identify data drifting and highlight areas with significant drift, we utilized meteorological air quality and weather data in this study. We proposed two models that consider the characteristics of data drifting for PM2.5 prediction and compared them with various deep learning models, such as Long Short-Term Memory (LSTM) and its variants, for predictions from the next hour to the 64th hour. Our proposed models significantly outperform traditional neural networks. Additionally, we introduced a wrapped loss function incorporated into a model, resulting in more accurate results compared to those using the original loss function alone and prediction has been evaluated by RMSE, MAE and MAPE metrics. The proposed Front-loaded connection model(FLC) and Back-loaded connection model (BLC) solve the data drifting issue and the wrap loss function also help alleviate the data drifting problem with model training and works for the neural network models to achieve more accurate results. Eventually, the experimental results have shown that the proposed model performance enhanced from 24.1% -16%, 12%-8.3% respectively at 1h-24h, 32h-64h with compared to baselines BILSTM model, by 24.6% -11.8%, 10%-10.2% respectively at 1h-24h, 32h-64h compared to CNN model in hourly PM2.5 predictions.

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.

PM2.5 concentration prediction based on EEMD-ALSTM

The concentration prediction of PM2.5 plays a vital role in controlling the air and improving the environment. This paper proposes a prediction model (namely EEMD-ALSTM) based on Ensemble Empirical Mode Decomposition (EEMD), Attention Mechanism and Long Short-Term Memory network (LSTM). Through the combination of decomposition and LSTM, attention mechanism is introduced to realize the prediction of PM2.5 concentration. The advantage of EEMD-ALSTM model is that it decomposes and combines the original data using the method of ensemble empirical mode decomposition, reduces the high nonlinearity of the original data, and Specially reintroduction the attention mechanism, which enhances the extraction and retention of data features by the model. Through experimental comparison, it was found that the EEMD-ALSTM model reduced its MAE and RMSE by about 15% while maintaining the same R2 correlation coefficient, and the stability of the model in the prediction process was also improved significantly.

The application of strategy based on LSTM for the short-term prediction of PM2.5 in city

Many cities have long suffered from the events of fine particulate matter (PM2.5) pollutions. The Taiwanese Government has long strived to accurately predict the short-term hourly concentration of PM2.5 for the warnings on air pollution. Long Short-Term Memory neural network (LSTM) based on deep learning improves the prediction accuracy of daily PM2.5 concentration but PM2.5 prediction for next hours still needs to be improved. Therefore, this study proposes innovative Application-Strategy-based LSTM (ASLSTM) to accurately predict the short-term hourly PM2.5 concentrations, especially for the high PM2.5 predictions. First, this study identified better spatiotemporal input feature of a LSTM for obtaining this Better LSTM (BLSTM). In doing so, BLSTM trained by appropriate datasets could accurately predict the next hourly pollution concentration. Next, the application strategy was applied on BLSTM to construct ASLSTM. Specifically, from a timeline perspective, ASLSTM concatenates several BLSTMs to predict the concentration of PM2.5 at the following next several hours during which the predicted outputs of BLSTM at this time t was selected and included as the inputs of the next BLSTM at the next time t + 1, and the oldest input used as BLSTM at the time t was removed. The result demonstrated that BLSTM were trained by the dataset collected from 2008 to 2010 at Dali measurement station because there is a relatively large amount of data on high PM2.5 concentration in this dataset. Besides, a comparison of the performance of the ASLSTM with that of the LSTM was made to validate this proposed ASLSTM, especially for the range of higher PM2.5 concentration that people concerned. More importantly, the feasibility of this proposed application strategy and the necessity of optimizing the input parameters of LSTM were validated. In summary, this ASLSTM could accurately predict the short-term PM2.5 in Taichung city.

Time series prediction of the chemical components of PM2.5 based on a deep learning model

Modeling-based prediction methods enable rapid, reagent-free air pollution detection based on inexpensive multi-source data than traditional chemical reaction-based detection methods in order to quickly understand the air pollution situation. In this study, a convolutional neural network (CNN) and long and short-term memory (LSTM) neural networks are integrated to create a CNN-LSTM time series prediction model to predict the concentration of PM2.5 and its chemical components (i.e., heavy metals, carbon component, and water-soluble ions) using meteorological data and air pollutants (PM2.5, SO2, NO2, CO, and O3). In the integrated CNN-LSTM model, the CNN uses convolutional and pooling layers to extract features from the data, whereas the powerful nonlinear mapping and learning capabilities of LSTM enable the time series prediction of air pollution. The experimental results showed that the CNN-LSTM exhibited good generalization ability in the prediction of As, Cd, Cr, Cu, Ni, and Zn, with a mean R2 above 0.9. Mean R2 predicted for PM2.5, Pb, Ti, EC, OC, SO42-, and NO3- ranged from 0.85 to 0.9. Shapley value showed that PM2.5, NO2, SO2, and CO had a greater influence on the predicted heavy metal results of the model. Regarding water-soluble ions, the predicted results were dominantly influenced by PM2.5, CO, and humidity. The prediction of the carbon fraction was affected mainly by the PM2.5 concentration. Additionally, several input variables for various components were eliminated without affecting the prediction accuracy of the model, with R2 between 0.70 and 0.84, thereby maximizing modeling efficiency and lowering operational costs. The fully trained model prediction results showed that most predicted components of PM2.5 were lower during January to March 2020 than those in 2018 and 2019. This study provides insight into improving the accuracy of modeling-based detection methods and promotes the development of integrated air pollution monitoring toward a more sustainable direction.

A nested machine learning approach to short-term PM2.5 prediction in metropolitan areas using PM2.5 data from different sensor networks

Many predictive models for ambient PM_2.5 concentrations rely on ground observations from a single monitoring network consisting of sparsely distributed sensors. Integrating data from multiple sensor networks for short-term PM_2.5 prediction remains largely unexplored. This paper presents a machine learning approach to predict ambient PM_2.5 concentration levels at any unmonitored location several hours ahead using PM_2.5 observations from nearby monitoring sites from two sensor networks and the location's social and environmental properties. Specifically, this approach first applies a Graph Neural Network and Long Short-Term Memory (GNN-LSTM) network to time series of daily observations from a regulatory monitoring network to make predictions of PM_2.5. This network produces feature vectors to store aggregated daily observations as well as dependency characteristics to predict daily PM_2.5. The daily feature vectors are then set as the precondition of the hourly level learning process. The hourly level learning again uses a GNN-LSTM network based on daily dependency information and hourly observations from a low-cost sensor network to produce spatiotemporal feature vectors capturing the combined dependency described by daily and hourly observations. Finally, the spatiotemporal feature vectors from the hourly learning process and social-environmental data are merged and used as the input to a single-layer Fully Connected (FC) network to output the predicted hourly PM_2.5 concentrations. To demonstrate the benefits of this novel prediction approach, we have conducted a case study using data collected from two sensor networks in Denver, CO, during 2021. Results show that the utilization of data from two sensor networks improves the overall performance of predicting fine-level, short-term PM_2.5 concentrations compared to other baseline models.

72-hour real-time forecasting of ambient PM2.5 by hybrid graph deep neural network with aggregated neighborhood spatiotemporal information

The observation-based air pollution forecasting method has high computational efficiency over traditional numerical models, but a poor ability in long-term (after 6 h) forecasting due to a lack of detailed representation of atmospheric processes associated with the pollution transport. To address such limitation, here we propose a novel real-time air pollution forecasting model that applies a hybrid graph deep neural network (GNN_LSTM) to dynamically capture the spatiotemporal correlations among neighborhood monitoring sites to better represent the physical mechanism of pollutant transport across the space with the graph structure which is established with features (angle, wind speed, and wind direction) of neighborhood sites to quantify their interactions. Such design substantially improves the model performance in 72-hour PM_2.5 forecasting over the whole Beijing-Tianjin-Hebei region (overall R² increases from 0.6 to 0.79), particularly for polluted episodes (PM_2.5 concentration > 55 µg/m³) with pronounced regional transport to be captured by GNN_LSTM model. The inclusion of the AOD feature further enhances the model performance in predicting PM_2.5 over the sites where the AOD can inform additional aloft PM_2.5 pollution features related to regional transport. The importance of neighborhood site (particularly for those in the upwind flow pathway of the target area) features for long-term PM_2.5 forecast is demonstrated by the increased performance in predicting PM_2.5 in the target city (Beijing) with the inclusion of additional 128 neighborhood sites. Moreover, the newly developed GNN_LSTM model also implies the "source"-receptor relationship, as impacts from distanced sites associated with regional transport grow along with the forecasting time (from 0% to 38% in 72 h) following the wind flow. Such results suggest the great potential of GNN_LSTM in long-term air quality forecasting and air pollution prevention.

Deciphering urban traffic impacts on air quality by deep learning and emission inventory

Air pollution is a major obstacle to future sustainability, and traffic pollution has become a large drag on the sustainable developments of future metropolises. Here, combined with the large volume of real-time monitoring data, we propose a deep learning model, iDeepAir, to predict surface-level PM_2.5 concentration in Shanghai megacity and link with MEIC emission inventory creatively to decipher urban traffic impacts on air quality. Our model exhibits high-fidelity in reproducing pollutant concentrations and reduces the MAE from 25.355 µg/m³ to 12.283 µg/m³ compared with other models. And identifies the ranking of major factors, local meteorological conditions have become a nonnegligible factor. Layer-wise relevance propagation (LRP) is used here to enhance the interpretability of the model and we visualize and analyze the reasons for the different correlation between traffic density and PM_2.5 concentration in various regions of Shanghai. Meanwhile, As the strict and effective industrial emission reduction measurements implementing in China, the contribution of urban traffic to PM_2.5 formation calculated by combining MEIC emission inventory and LRP is gradually increasing from 18.03% in 2011 to 24.37% in 2017 in Shanghai, and the impact of traffic emissions would be ever-prominent in 2030 according to our prediction. We also infer that the promotion of vehicular electrification would achieve further alleviation of PM_2.5 about 8.45% by 2030 gradually. These insights are of great significance to provide the decision-making basis for accurate and high-efficient traffic management and urban pollution control, and eventually benefit people's lives and high-quality sustainable developments of cities.

Deep learning hybrid predictions for the amount of municipal solid waste: A case study in Shanghai

It is crucial to precisely estimate the municipal solid waste (MSW) amount for its sustainable management. Owing to learning complicated and abstract features between the factors and target, deep learning has recently emerged as one of the useful tools with potential to predict the MSW amount. Therefore, this study aimed to design an MSW amount predicted system in Shanghai, consisting of Attention (A), one-dimensional convolutional neural network (C), and long short-term memory (L), to investigate the relationship between exogenous series (24 socioeconomics factors and past MSW amount) and target (MSW amount). The role of Attention, 1D-CNN, LSTM played on the MSW predicted amount also have investigated. The results show that attention is crucial for decoding the encoding information, which would improve performance between predicted and known MSW amount (R² in A-L-C, L-A-C, L-C-A was 89.45%, 90.77%, and 95.31%, respectively.). CNN modules appear to be positioned similarly across the MSW predicted system. Finally, R² in L-A-C, A-L-C, and A-C-L was 85.44%, 91.61%, and 89.45%, which suggested that LSTM as an intermediary between CNN and Attention modules seems a wise measure to predict the MSW amount based on the correlation efficiency. In addition, some socioeconomic factors including the average number of people in households and budget revenue may be chosen for the decision-making of MSW management in Shanghai city in the future, according to the weight of neurons in fully connected layers by the visual technology.

Forecasting Fine Particulate Matter Concentrations by In-Depth Learning Model According to Random Forest and Bilateral Long- and Short-Term Memory Neural Networks

Accurate prediction of fine particulate matter concentration in the future is important for human health due to the necessity of an early warning system. Generally, deep learning methods, when widely used, perform better in forecasting the concentration of PM2.5. However, the source information is limited, and the dynamic process is uncertain. The method of predicting short-term (3 h) and long-term trends has not been achieved. In order to deal with the issue, the research employed a novel mixed forecasting model by coupling the random forest (RF) variable selection and bidirectional long- and short-term memory (BiLSTM) neural net in order to forecast concentrations of PM2.5/0~12 h. Consequently, the average absolute percentage error of 1, 6, and 12 h shows that the PM2.5 concentration prediction is 3.73, 9.33, and 12.68 μg/m3 for Beijing, 1.33, 3.38, and 4.60 μg/m3 for Guangzhou, 1.37, 4.19, and 6.35 μg/m3 for Xi’an, and 2.20, 7.75, and 10.07 μg/m3 for Shenyang, respectively. Moreover, the results show that the suggested mixed model is an advanced method that can offer high accuracy of PM2.5 concentrations from 1 to 12 h post.

Research on PM2.5 Concentration Prediction Based on the CE-AGA-LSTM Model

The PM2.5 index is an important basis for measuring the degree of air pollution. The accurate prediction of PM2.5 concentration has an important guiding role in air pollution prevention and control. The Pearson Correlation Coefficient (PCC) is a common index used to mine the correlation between meteorological factors and other air pollutants. However, this index cannot be used to mine non-linear correlations, nor can it quantitatively analyze the weight of each related attribute. In order to accurately explore the correlation between meteorological factors and other air pollutants and to achieve an accurate prediction of PM2.5 concentration, this paper proposes a short- and long-time memory (LSTM) network prediction model based on Copula entropy (CE) and the adaptive genetic algorithm (AGA). By calculating CE, the correlation between multiple meteorological factors and various atmospheric pollutants and PM2.5 was analyzed. The correlation of influencing factors was sorted according to the size of the correlation coefficients. The contribution rate of meteorological factors and atmospheric pollutants to PM2.5 concentration was determined, used as the weight of each influencing factor and predicted as the input data of the prediction model. In this paper, a long- and short-term memory network (LSTM) suitable for time series data was selected as the prediction model, while the selection of model parameters was taken into account, and the relevant parameters were sought by an adaptive genetic algorithm (AGA). The air pollutant data and meteorological data of Beijing from 1 January 2016 to 31 December 2016 were selected, and MAE and RMSE were used as evaluation indexes. By comparing the experimental results of the CE-AGA-LSTM with those of other eight prediction models (LR, SVM, RF, ARMA, ST-LSTM, LSTM, CE-LSTM and CE-RNN), we found that among the models, the CE-AGA-LSTM model provided the lowest MAE and RMSE values, i.e., 14.5 and 21.88, respectively. At the same time, the loss rate and accuracy of the CE-AGA-LSTM model were evaluated, and the experimental results verified the validity of the model.

An air quality index prediction model based on CNN-ILSTM

Air quality index (AQI) is an essential measure of air pollution evaluation, which describes the air pollution degree and its impact on health, so the accurate prediction of AQI is significant. This paper presents an AQI prediction model based on Convolution Neural Networks (CNN) and Improved Long Short-Term Memory (ILSTM), named CNN-ILSTM. ILSTM deletes the output gate in LSTM and improves its input gate and forget gate, and introduces a Conversion Information Module (CIM) to prevent supersaturation in the learning process. ILSTM realizes efficient learning of historical data, improves prediction accuracy, and reduces the training time. CNN extracts the eigenvalues of input data effectively. This paper uses air quality data from 00:00 on January 1, 2017, to 23:00 on June 30, 2021, in Shijiazhuang City, Hebei Province, China, as experimental data sets, and compares this model with eight prediction models: SVR, RFR, MLP, LSTM, GRU, ILSTM, CNN-LSTM, and CNN-GRU to prove the validity and accuracy of CNN-ILSTM prediction model. The experimental results show the MAE of CNN-ILSTM is 8.4134, MSE is 202.1923, R2 is 0.9601, and the training time is 85.3 s. In this experiment, the performance of this model performs better than other models.

24-Hour prediction of PM2.5 concentrations by combining empirical mode decomposition and bidirectional long short-term memory neural network

Accurate prediction of the future PM2.5 concentration is crucial to human health and ecological environmental protection. Nowadays, deep learning methods show advantages in the prediction of PM2.5 concentration, but few of the studies can achieve accurate prediction of short term (within 6 h) concentration and also catch longer term (6-24 h) change trends. To address this issue, this study constructs a novel hybrid prediction model by combining the empirical mode decomposition (EMD) method, sample entropy (SE) index and bidirectional long and short-term memory neural network (BiLSTM) to predict 0-24 h PM2.5 concentrations. The experimental results show that the hybrid model has good performance on PM2.5 prediction with R2 = 0.987, RMSE = 2.77 μg/m3 at T + 1 moment and R2 = 0.904, RMSE = 7.51 μg/m3 at T + 6 moment. The novel model improves the accuracy on short-term (within 6 h) prediction of PM2.5 concentrations by at least 50% compared with other single deep learning models. Moreover, it well catches the variation trend of PM2.5 concentrations after 6 h till 24 h.

A balanced social LSTM for PM2.5 concentration prediction based on local spatiotemporal correlation

Reliable prediction for the concentration of PM_2.5 has become a hot topic in pollution prevention. However, the prediction for PM_2.5 concentration remains a challenge, one of the reasons is that current prediction methods do not consider the relevance of PM_2.5 concentration among surrounding areas. In this paper, we propose the assumption that the PM_2.5 concentration has spatial interaction, which includes two parts: 1) The PM_2.5 concentrations observed by adjacent stations usually present relevant trends; 2) Stations with higher PM_2.5 concentration tend to show higher influences on neighboring areas. Based on the spatial interaction assumption, we propose a balanced social long short-term memory (BS-LSTM) neural network for the prediction of PM_2.5 concentration. BS-LSTM is composed of two kernel components: a social-LSTM based prediction model and a new balanced mean squared error (B-MSE) based loss function. On the one hand, to capture the spatiotemporal correlation of the PM_2.5 concentration among adjacent stations, we develop a social-LSTM based model which has advantages in describing the trend information of neighboring locations. On the other hand, considering the unbalanced influence caused by various local pollution levels, we design a new B-MSE loss function to assign different attention to the observation stations. In the experiments, we evaluate the proposed method on two real-world PM_2.5 datasets. The results indicate that BS-LSTM is promising, especially in the case of heavy pollution.

A systematic literature review of deep learning neural network for time series air quality forecasting

Rapid progress of industrial development, urbanization and traffic has caused air quality reduction that negatively affects human health and environmental sustainability, especially among developed countries. Numerous studies on the development of air quality forecasting model using machine learning have been conducted to control air pollution. As such, there are significant numbers of reviews on the application of machine learning in air quality forecasting. Shallow architectures of machine learning exhibit several limitations and yield lower forecasting accuracy than deep learning architecture. Deep learning is a new technology in computational intelligence; thus, its application in air quality forecasting is still limited. This study aims to investigate the deep learning applications in time series air quality forecasting. Owing to this, literature search is conducted thoroughly from all scientific databases to avoid unnecessary clutter. This study summarizes and discusses different types of deep learning algorithms applied in air quality forecasting, including the theoretical backgrounds, hyperparameters, applications and limitations. Hybrid deep learning with data decomposition, optimization algorithm and spatiotemporal models are also presented to highlight those techniques' effectiveness in tackling the drawbacks of individual deep learning models. It is clearly stated that hybrid deep learning was able to forecast future air quality with higher accuracy than individual models. At the end of the study, some possible research directions are suggested for future model development. The main objective of this review study is to provide a comprehensive literature summary of deep learning applications in time series air quality forecasting that may benefit interested researchers for subsequent research.