Probabilistic concentration prediction of PM_2.5 in subway stations based on multi-resolution elastic-gated attention mechanism and Gaussian mixture model

Abstract

The subway has increasingly taken over as the primary method of short-distance travel in the development of modern urban transportation. Due to the poor air mobility in subway stations, it is crucial to monitor and provide alerts on air quality. This work provides a probabilistic prediction framework of PM_2.5 concentration to solve the air quality early warning problem in subway stations. Firstly, outliers are discovered and corrected utilizing a probabilistic-based auto-encoder (PAE). Secondly, the multi-resolution elastic-gated attention mechanism is used to address error accumulation and historical information lost during the prediction process. Moreover, the decoder structure of sequence to sequence (Seq2Seq) is improved through multiple output strategy and flexible gate attention mechanism to reduce error accumulation. Finally, the Seq2Seq is equipped with Gaussian mixture models (GMM), allowing it to adapt to more complicated changes and produce a probability distribution of PM_2.5 concentrations. According to tests on data from subway stations, the Pinball loss and Winkler score of the proposed model are smaller compared to other models.

Keywords

PM_2.5 / subway station / mixed Gaussian distribution / probabilistic prediction / attention mechanism

References

[1] SonYS, OhY H, ChoiI Y, et al. . Development of a magnetic hybrid filter to reduce PM₁₀ in a subway platform [J]. Journal of Hazardous Materials, 2019, 368: 197-203

[2] XiaoD, LiB, ChengS. The effect of subway development on air pollution: Evidence from China [J]. Journal of Cleaner Production, 2020, 275: 124149

[3] ZhengX, LiH, MengL, et al. . Improved social force model based on exit selection for microscopic pedestrian simulation in subway station [J]. Journal of Central South University, 2015, 22(11): 4490-4497

[4] SmithJ D, BarrattB M, FullerG W, et al. . PM_2.5 on the London underground [J]. Environment International, 2020, 134105188

[5] PanS, DuS, WangX, et al. . Analysis and interpretation of the particulate matter (PM₁₀ and PM_2.5) concentrations at the subway stations in Beijing, China [J]. Sustainable Cities and Society, 2019, 45366-377

[6] ParchamiM, MotlaghM S, PardakhtiA. Particulate matter concentrations and characterization in urban subway system-case study Tehran, Iran [J]. Journal of Environmental Health Science and Engineering, 2022, 20(2): 953-962

[7] QiaoT, XiuG, ZhengY, et al. . Preliminary investigation of PM₁, PM_2.5, PM₁₀ and its metal elemental composition in tunnels at a subway station in Shanghai, China [J]. Transportation Research Part D: Transport and Environment, 2015, 41: 136-146

[8] LiZ, CheW, FreyH C, et al. . Factors affecting variability in PM_2.5 exposure concentrations in a metro system [J]. Environmental Research, 2018, 160: 20-26

[9] Figueroa-LaraJ J, Murcia-GonzálezJ M, García-MartínezR, et al. . Effect of platform subway depth on the presence of Airborne PM_2.5, metals, and toxic organic species [J]. Journal of Hazardous Materials, 2019, 377: 427-436

[10] Mugica-ÁlvarezV, Figueroa-LaraJ, Romero-RomoM, et al. . Concentrations and properties of airborne particles in the Mexico City subway system [J]. Atmospheric Environment, 2012, 49: 284-293

[11] GusevaC I, CrézéC, HemmendingerM, et al. . Particle and metal exposure in Parisian subway: Relationship between exposure biomarkers in air, exhaled breath condensate, and urine [J]. International Journal of Hygiene and Environmental Health, 2021, 237: 113837

[12] GuoL, HuY, HuQ, et al. . Characteristics and chemical compositions of particulate matter collected at the selected metro stations of Shanghai, China [J]. Science of the Total Environment, 2014, 496443-452

[13] ZhangM, ShaoL, JonesT, et al. . Atmospheric iron particles in PM_2.5 from a subway station, Beijing, China [J]. Atmospheric Environment, 2022, 283: 119175

[14] JiW, LiX, WangC. Composition and exposure characteristics of PM_2.5 on subway platforms and estimates of exposure reduction by protective masks [J]. Environmental Research, 2021, 197111042

[15] RoyD, LyouE S, KimJ, et al. . Commuters health risk associated with particulate matter exposures in subway system-Globally [J]. Building and Environment, 2022, 216109036

[16] KimM, SankararaoB, KangO, et al. . Monitoring and prediction of indoor air quality (IAQ) in subway or metro systems using season dependent models [J]. Energy and Buildings, 2012, 4648-55

[17] LimJ, KimY, OhT, et al. . Analysis and prediction of indoor air pollutants in a subway station using a new key variable selection method [J]. Korean Journal of Chemical Engineering, 2012, 29(8): 994-1003

[18] LiuH, YangC, HuangM, et al. . Modeling of subway indoor air quality using Gaussian process regression [J]. Journal of Hazardous Materials, 2018, 359: 266-273

[19] FuM, WangW, LeZ, et al. . Prediction of particular matter concentrations by developed feed-forward neural network with rolling mechanism and gray model [J]. Neural Computing and Applications, 2015, 26(8): 1789-1797

[20] BalramD, LianK Y, SebastianN. Air quality warning system based on a localized PM_2.5 soft sensor using a novel approach of Bayesian regularized neural network via forward feature selection [J]. Ecotoxicology and Environmental Safety, 2019, 182: 109386

[21] GarcíaN P J, SánchezL F, García-GonzaloE, et al. . PM₁₀ concentration forecasting in the metropolitan area of Oviedo (Northern Spain) using models based on SVM, MLP, VARMA and ARIMA: A case study [J]. Science of the Total Environment, 2018, 621753-761

[22] DongM, YangD, KuangY, et al. . PM_2.5 concentration prediction using hidden semi-Markov model-based times series data mining [J]. Expert Systems With Applications, 2009, 36(5): 9046-9055

[23] SunW, ZhangH, PalazogluA, et al. . Prediction of 24-hour-average PM_2.5 concentrations using a hidden Markov model with different emission distributions in Northern California [J]. Science of the Total Environment, 2013, 443: 93-103

[24] WangJ, LiH, LuH. Application of a novel early warning system based on fuzzy time series in urban air quality forecasting in China [J]. Applied Soft Computing, 2018, 71: 783-799

[25] KarimianH, LiQ, WuC, et al. . Evaluation of different machine learning approaches to forecasting PM_2.5 mass concentrations [J]. Aerosol and Air Quality Research, 2019, 19(6): 1400-1410

[26] KimB Y, LimY K, ChaJ W. Short-term prediction of particulate matter (PM₁₀ and PM_2.5) in Seoul, South Korea using tree-based machine learning algorithms [J]. Atmospheric Pollution Research, 2022, 13(10): 101547

[27] LiuH, DengD. An enhanced hybrid ensemble deep learning approach for forecasting daily PM_2.5 [J]. Journal of Central South University, 2022, 2962074-2083

[28] Loy-BenitezJ, TariqS, NguyenH T, et al. . Neural circuit policies-based temporal flexible soft-sensor modeling of subway PM_2.5 with applications on indoor air quality management [J]. Building and Environment, 2022, 207: 108537

[29] PakU, MaJ, RyuU, et al. . Deep learning-based PM_2.5 prediction considering the spatiotemporal correlations: A case study of Beijing, China [J]. Science of the Total Environment, 2020, 699133561

[30] ZhangB, ZhangH, ZhaoG, et al. . Constructing a PM_2.5 concentration prediction model by combining auto-encoder with Bi-LSTM neural networks [J]. Environmental Modelling & Software, 2020, 124104600

[31] SohP W, ChangJ, HuangJ W. Adaptive deep learning-based air quality prediction model using the most relevant spatial-temporal relations [J]. IEEE Access, 2018, 638186-38199

[32] YeoI, ChoiY, LopsY, et al. . Efficient PM_2.5 forecasting using geographical correlation based on integrated deep learning algorithms [J]. Neural Computing and Applications, 2021, 33(22): 15073-15089

[33] ZhangL, LiuP, ZhaoL, et al. . Air quality predictions with a semi-supervised bidirectional LSTM neural network [J]. Atmospheric Pollution Research, 2021, 12(1): 328-339

[34] YangB, SunS, LiJ, et al. . Traffic flow prediction using LSTM with feature enhancement [J]. Neurocomputing, 2019, 332: 320-327

[35] BAHDANAU D, CHO K, BENGIO Y. Neural machine translation by jointly learning to align and translate [EB/OL]. 2014. arXiv: 1409.0473. https://arxiv.org/abs/1409.0473.

[36] BI Jing, LIN Yong-ze, DONG Quan-xi, et al. An improved attention-based LSTM for multi-step dissolved oxygen prediction in water environment [C]// 2020 IEEE International Conference on Networking, Sensing and Control (ICNSC). IEEE, 2020: 1–6. DOI: https://doi.org/10.1109/ICNSC48988.2020.9238097.

[37] XueL, LiX, ZhangN L. Not all attention is needed: Gated attention network for sequence data [J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(4): 6550-6557

[38] WangZ, ChenL, DingZ, et al. . An enhanced interval PM_2.5 concentration forecasting model based on BEMD and MLP^I with influencing factors [J]. Atmospheric Environment, 2020, 223: 117200

[39] ZhouM, WangB, GuoS, et al. . Multi-objective prediction intervals for wind power forecast based on deep neural networks [J]. Information Sciences, 2021, 550207-220

[40] JiangP, LiC, LiR, et al. . An innovative hybrid air pollution early-warning system based on pollutants forecasting and Extenics evaluation [J]. Knowledge-Based Systems, 2019, 164: 174-192

[41] LiuH, DuanZ, ChenC. A hybrid framework for forecasting PM_2.5 concentrations using multi-step deterministic and probabilistic strategy [J]. Air Quality, Atmosphere & Health, 2019, 12(7): 785-795

[42] YongN K, AwangN. Wavelet-based time series model to improve the forecast accuracy of PM₁0 concentrations in Peninsular Malaysia [J]. Environmental Monitoring and Assessment, 2019, 191(2): 1-12

[43] LiuH, DuanZ, ChenC. A hybrid multiresolution multi-objective ensemble model and its application for forecasting of daily PM_2.5 concentrations [J]. Information Sciences, 2020, 516: 266-292

[44] SalinasD, FlunkertV, GasthausJ, et al. . DeepAR: Probabilistic forecasting with autoregressive recurrent networks [J]. International Journal of Forecasting, 2020, 36(3): 1181-1191

[45] LiG, ChenL, YangH. Prediction of PM_2.5 concentration based on improved secondary decomposition and CSA-KELM [J]. Atmospheric Pollution Research, 2022, 13(7): 101455

[46] PinsonP, TastuJ. Discussion of “prediction intervals for short-term wind farm generation forecasts” and “combined nonparametric prediction intervals for wind power generation” [J]. IEEE Transactions on Sustainable Energy, 2014, 531019-1020

[47] NowotarskiJ, WeronR. Recent advances in electricity price forecasting: A review of probabilistic forecasting [J]. Renewable and Sustainable Energy Reviews, 2018, 81: 1548-1568

[48] HarrL, SinselT, SimonH, et al. . PM_2.5 exposure differences between children and adults [J]. Urban Climate, 2022, 44101198

[49] XuC, ChenH. A hybrid data mining approach for anomaly detection and evaluation in residential buildings energy data [J]. Energy and Buildings, 2020, 215109864

[50] JiangF, ZhuQ, TianT. An ensemble interval prediction model with change point detection and interval perturbation-based adjustment strategy: A case study of air quality [J]. Expert Systems With Applications, 2023, 222119823

[51] YaoW, ZengZ, LianC. Generating probabilistic predictions using mean-variance estimation and echo state network [J]. Neurocomputing, 2017, 219(C): 536-547