1
|
Quantifying uncertainty: Air quality forecasting based on dynamic spatial-temporal denoising diffusion probabilistic model. ENVIRONMENTAL RESEARCH 2024; 249:118438. [PMID: 38350546 DOI: 10.1016/j.envres.2024.118438] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Received: 01/14/2024] [Revised: 02/04/2024] [Accepted: 02/05/2024] [Indexed: 02/15/2024]
Abstract
Air pollution constitutes a substantial peril to human health, thereby catalyzing the evolution of an array of air quality prediction models. These models span from mechanistic and statistical strategies to machine learning methodologies. The burgeoning field of deep learning has given rise to a plethora of advanced models, which have demonstrated commendable performance. However, previous investigations have overlooked the salience of quantifying prediction uncertainties and potential future interconnections among air monitoring stations. Moreover, prior research typically utilized static predetermined spatial relationships, neglecting dynamic dependencies. To address these limitations, we propose a model named Dynamic Spatial-Temporal Denoising Diffusion Probabilistic Model (DST-DDPM) for air quality prediction. Our model is underpinned by the renowned denoising diffusion model, aiding us in discerning indeterminacy. In order to encapsulate dynamic patterns, we design a dynamic context encoder to generate dynamic adjacency matrices, whilst maintaining static spatial information. Furthermore, we incorporate a spatial-temporal denoising model to concurrently learn both spatial and temporal dependencies. Authenticating our model's performance using a real-world dataset collected in Beijing, the outcomes indicate that our model eclipses other baseline models in terms of both short-term and long-term predictions by 1.36% and 11.62% respectively. Finally, we conduct a case study to exhibit our model's capacity to quantify uncertainties.
Collapse
|
2
|
Time-sensitive prediction of NO 2 concentration in China using an ensemble machine learning model from multi-source data. J Environ Sci (China) 2024; 137:30-40. [PMID: 37980016 DOI: 10.1016/j.jes.2023.02.026] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 10/31/2022] [Revised: 02/02/2023] [Accepted: 02/13/2023] [Indexed: 11/20/2023]
Abstract
Nitrogen dioxide (NO2) poses a critical potential risk to environmental quality and public health. A reliable machine learning (ML) forecasting framework will be useful to provide valuable information to support government decision-making. Based on the data from 1609 air quality monitors across China from 2014-2020, this study designed an ensemble ML model by integrating multiple types of spatial-temporal variables and three sub-models for time-sensitive prediction over a wide range. The ensemble ML model incorporates a residual connection to the gated recurrent unit (GRU) network and adopts the advantage of Transformer, extreme gradient boosting (XGBoost) and GRU with residual connection network, resulting in a 4.1%±1.0% lower root mean square error over XGBoost for the test results. The ensemble model shows great prediction performance, with coefficient of determination of 0.91, 0.86, and 0.77 for 1-hr, 3-hr, and 24-hr averages for the test results, respectively. In particular, this model has achieved excellent performance with low spatial uncertainty in Central, East, and North China, the major site-dense zones. Through the interpretability analysis based on the Shapley value for different temporal resolutions, we found that the contribution of atmospheric chemical processes is more important for hourly predictions compared with the daily scale predictions, while the impact of meteorological conditions would be ever-prominent for the latter. Compared with existing models for different spatiotemporal scales, the present model can be implemented at any air quality monitoring station across China to facilitate achieving rapid and dependable forecast of NO2, which will help developing effective control policies.
Collapse
|
3
|
An adaptive adjacency matrix-based graph convolutional recurrent network for air quality prediction. Sci Rep 2024; 14:4408. [PMID: 38388632 PMCID: PMC10883962 DOI: 10.1038/s41598-024-55060-2] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/01/2023] [Accepted: 02/20/2024] [Indexed: 02/24/2024] Open
Abstract
In recent years, air pollution has become increasingly serious and poses a great threat to human health. Timely and accurate air quality prediction is crucial for air pollution early warning and control. Although data-driven air quality prediction methods are promising, there are still challenges in studying spatial-temporal correlations of air pollutants to design effective predictors. To address this issue, a novel model called adaptive adjacency matrix-based graph convolutional recurrent network (AAMGCRN) is proposed in this study. The model inputs Point of Interest (POI) data and meteorological data into a fully connected neural network to learn the weights of the adjacency matrix thereby constructing the self-ringing adjacency matrix and passes the pollutant data with this matrix as input to the Graph Convolutional Network (GCN) unit. Then, the GCN unit is embedded into LSTM units to learn spatio-temporal dependencies. Furthermore, temporal features are extracted using Long Short-Term Memory network (LSTM). Finally, the outputs of these two components are merged and air quality predictions are generated through a hidden layer. To evaluate the performance of the model, we conducted multi-step predictions for the hourly concentration of PM2.5, PM10 and O3 at Fangshan, Tiantan and Dongsi monitoring stations in Beijing. The experimental results show that our method achieves better predicted effects compared with other baseline models based on deep learning. In general, we designed a novel air quality prediction method and effectively addressed the shortcomings of existing studies in learning the spatio-temporal correlations of air pollutants. This method can provide more accurate air quality predictions and is expected to provide support for public health protection and government environmental decision-making.
Collapse
|
4
|
An interval AQI combination prediction model based on multiple data decomposition and information aggregation operator. ENVIRONMENTAL SCIENCE AND POLLUTION RESEARCH INTERNATIONAL 2024; 31:14284-14302. [PMID: 38277105 DOI: 10.1007/s11356-024-32060-3] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Received: 08/21/2023] [Accepted: 01/15/2024] [Indexed: 01/27/2024]
Abstract
In this paper, an interval Air Quality Index (AQI) combination prediction model based on EEMD, VMD, and the weighted power average (WPA) operator is proposed. EEMD and VMD decompose complex AQI data effectively, while WPA operator reasonably aggregates the prediction results of different models. We validate the effectiveness of the proposed model using Shenzhen's daily interval AQI. Furthermore, three kinds of prediction models are compared with the proposed model to highlight its advantages from various perspectives. The results show that the introduction of data decomposition methods significantly improves the model's prediction accuracy, WPA operator further enhances the model's prediction capability, and the incorporation of EEMD and VMD enables the proposed model to have stronger feature extraction capabilities for complex time series. As a result, the model proposed in this paper demonstrates strong generalization ability and prediction accuracy, making it applicable not only for air quality prediction but also for other domains such as economics and environment.
Collapse
|
5
|
Air quality prediction by integrating mechanism model and machine learning model. THE SCIENCE OF THE TOTAL ENVIRONMENT 2023; 899:165646. [PMID: 37474048 DOI: 10.1016/j.scitotenv.2023.165646] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Subscribe] [Scholar Register] [Received: 04/24/2023] [Revised: 06/26/2023] [Accepted: 07/17/2023] [Indexed: 07/22/2023]
Abstract
AQP (Air Quality Prediction) is a very challenging project, and its core issue is how to solve the interaction and influence among meteorological, spatial and temporal factors. To address this central conundrum, we make full use of the characteristics of mechanism model and machine learning and propose a new AQP method based on DM_STGNN (Dynamic Multi-granularity Spatio-temporal Graph Neural Network). This method is the first time to use the air quality model HYSPLIT (Hybrid Single-Particle Lagrangian Integrated Trajectory Model) to assist in building a dynamic spatio-temporal graph structure to learn the spatiotemporal relationship of pollutants. DM_STGNN is based on an elaborate encoder-decoder architecture. At the encoder, in order to better mine the spatial dependency, we built a multi-granularity graph structure, used meteorological, time and geographical features to establish node attributes, used well-known HYSPLIT model to dynamically establish the edges among nodes, and used LSTM (Long Short Term Memory) to learn the time-series relationship of pollutant concentrations. At the decoder, in order to better mine the temporal dependency, we built an attention mechanism based LSTM for decoding and AQP. Additionally, in order to efficiently learn the temporal patterns from very long-term historical time series and generate rich contextual information, an unsupervised pre-training model is used to enhance DM_STGNN. The proposed model makes full use of and fully considers the influence of meteorological, spatial and temporal factors, and integrates the advantages of mechanism model and machine learning. On a project-based dataset, we validate the effectiveness of the proposed model and examine its abilities of capturing both fine-grained and long-term influences in AQP. We also compare the proposed model with the state-of-the-art AQP methods on the dataset of Yangtze River Delta city group, the experimental results show the appealing performance of our model over competitive baselines.
Collapse
|
6
|
Dynamic graph neural network with adaptive edge attributes for air quality prediction: A case study in China. Heliyon 2023; 9:e17746. [PMID: 37456022 PMCID: PMC10345359 DOI: 10.1016/j.heliyon.2023.e17746] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 07/09/2022] [Revised: 06/27/2023] [Accepted: 06/27/2023] [Indexed: 07/18/2023] Open
Abstract
Air quality prediction is a typical Spatiotemporal modeling problem, which always uses different components to handle spatial and temporal dependencies in complex systems separately. Previous models based on time series analysis and recurrent neural network (RNN) methods have only modeled time series while ignoring spatial information. Previous graph convolution neural networks (GCNs) based methods usually require providing spatial correlation graph structure of observation sites in advance. The correlations among these sites and their strengths are usually calculated using prior information. However, due to the limitations of human cognition, limited prior information cannot reflect the real station-related structure or bring more effective information for accurate prediction. To this end, we propose a novel Dynamic Graph Neural Network with Adaptive Edge Attributes (DGN-AEA) on the message passing network, which generates the adaptive bidirected dynamic graph by learning the edge attributes as model parameters. Unlike prior information to establish edges, our method can obtain adaptive edge information through end-to-end training without any prior information. Thus reducing the complexity of the problem. Besides, the hidden structural information between the stations can be obtained as model by-products, which can help make some subsequent decision-making analyses. Experimental results show that our model received state-of-the-art performance than other baselines.
Collapse
|
7
|
Multi-view multi-task spatiotemporal graph convolutional network for air quality prediction. THE SCIENCE OF THE TOTAL ENVIRONMENT 2023:164699. [PMID: 37315618 DOI: 10.1016/j.scitotenv.2023.164699] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Subscribe] [Scholar Register] [Received: 12/01/2022] [Revised: 04/18/2023] [Accepted: 06/04/2023] [Indexed: 06/16/2023]
Abstract
Accurate air quality prediction is a crucial but arduous task for intelligent cities. Predictable air quality can advise governments on environmental governance and residents on travel. However, complex correlations (i.e., intra-sensor correlation and inter-sensor correlation) make prediction challenging. Previous work considered the spatial, temporal, or combination of the two to model. However, we observe that there are also logical semantic and temporal, and spatial relations. Therefore, we propose a multi-view multi-task spatiotemporal graph convolutional network (M2) for air quality prediction. We encode three views, including spatial view (using GCN to model the correlation between adjacent stations in geographic space), logical view (using GCN to model the correlation between stations in logical space), and temporal view (using GRU to model the correlation among historical data). Meanwhile, M2 chooses a multi-task learning paradigm that includes a classification task (auxiliary task, coarse granularity prediction of air quality level) and a regression task (main task, fine granularity prediction of air quality value) to predict jointly. And the experimental results on two real-world air quality datasets demonstrate our model performances over the state-of-art methods.
Collapse
|
8
|
Machine learning algorithms for high-resolution prediction of spatiotemporal distribution of air pollution from meteorological and soil parameters. ENVIRONMENT INTERNATIONAL 2023; 175:107931. [PMID: 37119651 DOI: 10.1016/j.envint.2023.107931] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Subscribe] [Scholar Register] [Received: 12/13/2022] [Revised: 03/18/2023] [Accepted: 04/11/2023] [Indexed: 05/22/2023]
Abstract
This study uses machine learning (ML) models for a high-resolution prediction (0.1°×0.1°) of air fine particular matter (PM2.5) concentration, the most harmful to human health, from meteorological and soil data. Iraq was considered the study area to implement the method. Different lags and the changing patterns of four European Reanalysis (ERA5) meteorological variables, rainfall, mean temperature, wind speed and relative humidity, and one soil parameter, the soil moisture, were used to select the suitable set of predictors using a non-greedy algorithm known as simulated annealing (SA). The selected predictors were used to simulate the temporal and spatial variability of air PM2.5 concentration over Iraq during the early summer (May-July), the most polluted months, using three advanced ML models, extremely randomized trees (ERT), stochastic gradient descent backpropagation (SGD-BP) and long short-term memory (LSTM) integrated with Bayesian optimizer. The spatial distribution of the annual average PM2.5 revealed the population of the whole of Iraq is exposed to a pollution level above the standard limit. The changes in temperature and soil moisture and the mean wind speed and humidity of the month before the early summer can predict the temporal and spatial variability of PM2.5 over Iraq during May-July. Results revealed the higher performance of LSTM with normalized root-mean-square error and Kling-Gupta efficiency of 13.4% and 0.89, compared to 16.02% and 0.81 for SDG-BP and 17.9% and 0.74 for ERT. The LSTM could also reconstruct the observed spatial distribution of PM2.5 with MapCurve and Cramer's V values of 0.95 and 0.91, compared to 0.9 and 0.86 for SGD-BP and 0.83 and 0.76 for ERT. The study provided a methodology for forecasting spatial variability of PM2.5 concentration at high resolution during the peak pollution months from freely available data, which can be replicated in other regions for generating high-resolution PM2.5 forecasting maps.
Collapse
|
9
|
Air quality prediction for Chengdu based on long short-term memory neural network with improved jellyfish search optimizer. ENVIRONMENTAL SCIENCE AND POLLUTION RESEARCH INTERNATIONAL 2023; 30:64416-64442. [PMID: 37067716 DOI: 10.1007/s11356-023-26782-z] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Received: 11/22/2022] [Accepted: 03/29/2023] [Indexed: 05/11/2023]
Abstract
Air quality prediction plays an important role in preventing air pollution and improving living environment. For this prediction, many indicators can be employed to reflect the air quality, among which air quality index (AQI) is the most commonly used. However, existing methods are relatively simple and the corresponding prediction accuracy needs to be improved. Particularly, the prediction accuracy is affected by the parameter selection of methods, and the corresponding optimization problems are usually non-convex and multi-modal. Therefore, based on long short-term memory (LSTM) neural network with improved jellyfish search optimizer (IJSO), a novel hybrid model denoted by IJSO-LSTM is proposed to predict AQI for Chengdu. In order to evaluate the optimizing ability of IJSO, other variants of jellyfish search optimizer as well as other state-of-the-art meta-heuristic algorithms are applied to optimize the hyperparameters of LSTM neural network for comparison, and the results confirm that IJSO is more suitable for optimizing LSTM neural network. In addition, compared with other well-known models, the results demonstrate IJSO-LSTM has higher prediction accuracy with root-mean-square error, mean absolute error, and mean absolute percentage error controlling below 4, 3, and 4%, respectively.
Collapse
|
10
|
Spatiotemporal neural network for estimating surface NO 2 concentrations over north China and their human health impact. ENVIRONMENTAL POLLUTION (BARKING, ESSEX : 1987) 2022; 307:119510. [PMID: 35605830 DOI: 10.1016/j.envpol.2022.119510] [Citation(s) in RCA: 7] [Impact Index Per Article: 3.5] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Received: 11/12/2021] [Revised: 05/16/2022] [Accepted: 05/17/2022] [Indexed: 06/15/2023]
Abstract
Atmospheric nitrogen dioxide (NO2) is an important reactive gas pollutant harmful to human health. The spatiotemporal coverage provided by traditional NO2 monitoring methods is insufficient, especially in the suburban and rural areas of north China, which have a high population density and experience severe air pollution. In this study, we implemented a spatiotemporal neural network (STNN) model to estimate surface NO2 from multiple sources of information, which included satellite and in situ measurements as well as meteorological and geographical data. The STNN predicted NO2 with high accuracy, with a coefficient of determination (R2) of 0.89 and a root mean squared error of 5.8 μg/m3 for sample-based 10-fold cross-validation. Based on the surface NO2 concentration determined by the STNN, we analyzed the spatial distribution and temporal trends of NO2 pollution in north China. We found substantial drops in surface NO2 concentrations ranging between 9.1% and 33.2% for large cities during the 2020 COVID-19 lockdown when compared to those in 2019. Moreover, we estimated the all-cause deaths attributed to NO2 exposure at a high spatial resolution of about 1 km, with totals of 6082, 4200, and 18,210 for Beijing, Tianjin, and Hebei Provinces in 2020, respectively. We observed remarkable regional differences in the health impacts due to NO2 among urban, suburban, and rural areas. Generally, the STNN model could incorporate spatiotemporal neighboring information and infer surface NO2 concentration with full coverage and high accuracy. Compared with machine learning regression techniques, STNN can effectively avoid model overfitting and simultaneously consider both spatial and temporal correlations of input variables using deep convolutional networks with residual blocks. The use of the proposed STNN model, as well as the surface NO2 dataset, can benefit air quality monitoring, forecasting, and health burden assessments.
Collapse
|
11
|
A dual-path dynamic directed graph convolutional network for air quality prediction. THE SCIENCE OF THE TOTAL ENVIRONMENT 2022; 827:154298. [PMID: 35271925 DOI: 10.1016/j.scitotenv.2022.154298] [Citation(s) in RCA: 10] [Impact Index Per Article: 5.0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Received: 10/31/2021] [Revised: 02/28/2022] [Accepted: 02/28/2022] [Indexed: 06/14/2023]
Abstract
Accurate air quality prediction can help cope with air pollution and improve the life quality. With the development of the deployments of low-cost air quality sensors, increasing data related to air quality has provided chances to find out more accurate prediction methods. Air quality is affected by many external factors such as the position, wind, meteorological information, and so on. Meanwhile, these factors are spatio-temporal dynamic and there are many dynamic contextual relationships between them. Many methods for air quality prediction do not consider these complex spatio-temporal correlations and dynamic contextual relationships. In this paper, we propose a dual-path dynamic directed graph convolutional network (DP-DDGCN) for air quality prediction. We first create a dual-path transposed dynamic directed graph according to static distance relationships of stations and the dynamic relationships generated by wind speed and directions. Then based on the dual-path dynamic directed graph, we can capture the dynamic spatial dependencies more comprehensively. After that we apply gated recurrent units (GRUs) and add the future meteorological features, to extract the complex temporal dependencies of historical air quality data. Using dual-path dynamic directed graph blocks and the GRUs, we finally construct a dynamic spatio-temporal gated recurrent block to capture the dynamic spatio-temporal contextual correlations. Based on real-world datasets, which record a large amount of PM2.5 concentration data, we compare the proposed model with the benchmark models. The experimental results show that our proposed model has the best performance in predicting the PM2.5 concentrations.
Collapse
|
12
|
A hybrid deep learning technology for PM 2.5 air quality forecasting. ENVIRONMENTAL SCIENCE AND POLLUTION RESEARCH INTERNATIONAL 2021; 28:39409-39422. [PMID: 33759095 DOI: 10.1007/s11356-021-12657-8] [Citation(s) in RCA: 18] [Impact Index Per Article: 6.0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Received: 10/23/2020] [Accepted: 01/20/2021] [Indexed: 06/12/2023]
Abstract
The concentration of PM2.5 is one of the main factors in evaluating the air quality in environmental science. The severe level of PM2.5 directly affects the public health, economics and social development. Due to the strong nonlinearity and instability of the air quality, it is difficult to predict the volatile changes of PM2.5 over time. In this paper, a hybrid deep learning model VMD-BiLSTM is constructed, which combines variational mode decomposition (VMD) and bidirectional long short-term memory network (BiLSTM), to predict PM2.5 changes in cities in China. VMD decomposes the original PM2.5 complex time series data into multiple sub-signal components according to the frequency domain. Then, BiLSTM is employed to predict each sub-signal component separately, which significantly improved forecasting accuracy. Through a comprehensive study with existing models, such as the EMD-based models and other VMD-based models, we justify the outperformance of the proposed VMD-BiLSTM model over all compared models. The results show that the prediction results are significantly improved with the proposed forecasting framework. And the prediction models integrating VMD are better than those integrating EMD. Among all the models integrating VMD, the proposed VMD-BiLSTM model is the most stable forecasting method.
Collapse
|
13
|
A feature selection and multi-model fusion-based approach of predicting air quality. ISA TRANSACTIONS 2020; 100:210-220. [PMID: 31812248 DOI: 10.1016/j.isatra.2019.11.023] [Citation(s) in RCA: 15] [Impact Index Per Article: 3.8] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Subscribe] [Scholar Register] [Received: 05/05/2019] [Revised: 11/15/2019] [Accepted: 11/17/2019] [Indexed: 06/10/2023]
Abstract
With the rapid development of China's industrialization, the air pollution is becoming more and more serious. It is vital for us to predict the air quality for determining the further prevention measures of avoiding the brought disasters. In this paper, we are going to propose an approach of predicting the air quality based on the multiple data features through fusing the multiple machine learning models. The approach takes the meteorological data and air quality data for the past six days as one batch of input (the whole data set is for 46 days) and employs a multi-model fusion to provide an improved 24-hour prediction of PM2.5 pollutant concentration all over Beijing. During the above process, two focal feature groups are composed. The first focal feature group contains the historical meteorological data, while the second group includes the statistical information, the date information and the polynomial variations. Besides the two groups, we complement one million more data items by employing the time sliding means. Among the supplementary data, we select the most critical 500 features with Light Gradient Boosting Machine (LightGBM) model and send the features as the input to Gradient Boosting Decision Tree (GBDT) and LightGBM models. Meanwhile, we screen the most critical 300 features with eXtreme Gradient Boosting (XGBoost) model and send them as the input to the three prediction models. Referring to each of the models, we respectively gain the optimal parameters through grid search methods and then fuse the models' contribution with the linear weighting. The experiments indicate that the proposed approach based on the weighting fusion is better than that provided by a single modeling scheme, and the loss value is 0.4158 under the SMAPE index.
Collapse
|
14
|
Air quality prediction at new stations using spatially transferred bi-directional long short-term memory network. THE SCIENCE OF THE TOTAL ENVIRONMENT 2020; 705:135771. [PMID: 31972931 DOI: 10.1016/j.scitotenv.2019.135771] [Citation(s) in RCA: 17] [Impact Index Per Article: 4.3] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Subscribe] [Scholar Register] [Received: 08/29/2019] [Revised: 11/22/2019] [Accepted: 11/24/2019] [Indexed: 06/10/2023]
Abstract
In the last decades, air pollution has been a critical environmental issue, especially in developing countries like China. The governments and scholars have spent lots of effort on controlling air pollution and mitigating its impacts on human society. Accurate prediction of air quality can provide essential decision-making supports, and therefore, scholars have proposed various kinds of models and methods for air quality forecastings, such as statistical methods, machine learning methods, and deep learning methods. Deep learning-based networks, such as RNN and LSTM, have been reported to achieve good performance in recent studies. However, the excellent performance of these methods requires sufficient data to train the model. For stations that lack data, such as newly built monitoring stations, the performance of those methods is constrained. Therefore, a methodology that could address the data shortage problem in new stations should be explored. This study proposes a transfer learning-based stacked bidirectional long short term memory (TLS-BLSTM) network to predict air quality for the new stations that lack data. The proposed method integrates advanced deep learning techniques and transfer learning strategies to transfer the knowledge learned from existing air quality stations to new stations to boost forecasting. A case study in Anhui, China, was conducted to evaluate the effectiveness of TLS-BLSTM. The results show that the proposed method can help achieve 35.21% lower RMSE on average for the experimented three pollutants in new stations.
Collapse
|
15
|
Deep learning architecture for air quality predictions. ENVIRONMENTAL SCIENCE AND POLLUTION RESEARCH INTERNATIONAL 2016; 23:22408-22417. [PMID: 27734318 DOI: 10.1007/s11356-016-7812-9] [Citation(s) in RCA: 47] [Impact Index Per Article: 5.9] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Received: 08/09/2016] [Accepted: 09/29/2016] [Indexed: 06/06/2023]
Abstract
With the rapid development of urbanization and industrialization, many developing countries are suffering from heavy air pollution. Governments and citizens have expressed increasing concern regarding air pollution because it affects human health and sustainable development worldwide. Current air quality prediction methods mainly use shallow models; however, these methods produce unsatisfactory results, which inspired us to investigate methods of predicting air quality based on deep architecture models. In this paper, a novel spatiotemporal deep learning (STDL)-based air quality prediction method that inherently considers spatial and temporal correlations is proposed. A stacked autoencoder (SAE) model is used to extract inherent air quality features, and it is trained in a greedy layer-wise manner. Compared with traditional time series prediction models, our model can predict the air quality of all stations simultaneously and shows the temporal stability in all seasons. Moreover, a comparison with the spatiotemporal artificial neural network (STANN), auto regression moving average (ARMA), and support vector regression (SVR) models demonstrates that the proposed method of performing air quality predictions has a superior performance.
Collapse
|