An enhanced hybrid ensemble deep learning approach for forecasting daily PM_2.5

Abstract

PM_2.5 forecasting technology can provide a scientific and effective way to assist environmental governance and protect public health. To forecast PM_2.5, an enhanced hybrid ensemble deep learning model is proposed in this research The whole framework of the proposed model can be generalized as follows: the original PM_2.5 series is decomposed into 8 sub-series with different frequency characteristics by variational mode decomposition (VMD); the long short-term memory (LSTM) network, echo state network (ESN), and temporal convolutional network (TCN) are applied for parallel forecasting for 8 different frequency PM_2.5 sub-series; the gradient boosting decision tree (GBDT) is applied to assemble and reconstruct the forecasting results of LSTM, ESN and TCN. By comparing the forecasting data of the models over 3 PM_2.5 series collected from Shenyang, Changsha and Shenzhen, the conclusions can be drawn that GBDT is a more effective method to integrate the forecasting result than traditional heuristic algorithms; MAE values of the proposed model on 3 PM_2.5 series are 1.587, 1.718 and 1.327 µg/m³, respectively and the proposed model achieves more accurate results for all experiments than sixteen alternative forecasting models which contain three state-of-the-art models.

Keywords

PM_2.5 forecasting / variational mode decomposition / deep neural network / ensemble learning

References

[1] GanK, SunS-L, WangS-Y, et al.. A secondary-decomposition-ensemble learning paradigm for forecasting PM_2.5 concentration [J]. Atmospheric Pollution Research, 2018, 9(6): 989-999

[2] KapposA D, BruckmannP, EikmannT, et al.. Health effects of particles in ambient air [J]. International Journal of Hygiene and Environmental Health, 2004, 207(4): 399-407

[3] LiangC-S, DuanF-K, HeK-B, et al.. Review on recent progress in observations, source identifications and countermeasures of PM_2.5 [J]. Environment International, 2016, 86: 150-170

[4] GrellG A, PeckhamS E, SchmitzR, et al.. Fully coupled “online” chemistry within the WRF model [J]. Atmospheric Environment, 2005, 39(37): 6957-6975

[5] DjalalovaI, Delle MonacheL, WilczakJ. PM_2.5 analog forecast and Kalman filter post-processing for the Community Multiscale Air Quality (CMAQ) model [J]. Atmospheric Environment, 2015, 119: 431-442

[6] HuangJ-P, McqueenJ, WilczakJ, et al.. Improving NOAA NAQFC PM_2.5 predictions with a bias correction approach [J]. Weather and Forecasting, 2017, 32(2): 407-421

[7] QuX, WangW, WangW-F, et al.. Real-time rear-end crash potential prediction on freeways [J]. Journal of Central South University, 2017, 24(11): 2664-2673

[8] SunW, SunJ-Y. Daily PM_2.5 concentration prediction based on principal component analysis and LSSVM optimized by cuckoo search algorithm [J]. Journal of Environmental Management, 2017, 188: 144-152

[9] AbdullahS, NapiN N L M, AhmedA N, et al.. Development of multiple linear regression for particulate matter (PM₁₀) forecasting during episodic transboundary haze event in Malaysia [J]. Atmosphere, 2020, 11(3): 289

[10] AgarwalS, SharmaS, R S, et al.. Air quality forecasting using artificial neural networks with real time dynamic error correction in highly polluted regions [J]. Science of the Total Environment, 2020, 735: 139454

[11] ArsovM, ZdravevskiE, LameskiP, et al.. Multi-horizon air pollution forecasting with deep neural networks [J]. Sensors, 2021, 2141235

[12] KalajdjieskiJ, ZdravevskiE, CorizzoR, et al.. Air pollution prediction with multi-modal data and deep neural networks [J]. Remote Sensing, 2020, 12244142

[13] BaiY, ZengB, LiC, et al.. An ensemble long short-term memory neural network for hourly PM_2.5 concentration forecasting [J]. Chemosphere, 2019, 222286-294

[14] XuY-N, LiuH, DuanZ. A novel hybrid model for multi-step daily AQI forecasting driven by air pollution big data [J]. Air Quality, Atmosphere & Health, 2020, 13(2): 197-207

[15] JIANG Fu-xin, ZHANG Cheng-yuan, SUN Shao-long, et al. A novel hybrid framework for hourly PM2.5 concentration forecasting using CEEMDAN and deep temporal convolutional neural network [OL] arXiv preprint [2020-12-07]. https://arxiv.orglabs/2012.03781.

[16] ChengY, ZhangH, LiuZ-H, et al.. Hybrid algorithm for short-term forecasting of PM_2.5 in China [J]. Atmospheric Environment, 2019, 200264-279

[17] LiuH, JinK-R, DuanZ. Air PM_2.5 concentration multi-step forecasting using a new hybrid modeling method: Comparing cases for four cities in China [J]. Atmospheric Pollution Research, 2019, 10(5): 1588-1600

[18] ZhouQ-P, JiangH-Y, WangJ-Z, et al.. A hybrid model for PM_2.5 forecasting based on ensemble empirical mode decomposition and a general regression neural network [J]. Science of the Total Environment, 2014, 496: 264-274

[19] HuangJ-H, LiuH. A hybrid decomposition-boosting model for short-term multi-step solar radiation forecasting with NARX neural network [J]. Journal of Central South University, 2021, 28(2): 507-526

[20] CaoH-R, FanF, ZhouK, et al.. Wheel-bearing fault diagnosis of trains using empirical wavelet transform [J]. Measurement, 2016, 82: 439-449

[21] LiuS, WangQ-D, LuoY-P. A review of applications of visual inspection technology based on image processing in the railway industry [J]. Transportation Safety and Environment, 2020, 1(3): 185-204

[22] WuQ-L, LinH-X. Daily urban air quality index forecasting based on variational mode decomposition, sample entropy and LSTM neural network [J]. Sustainable Cities and Society, 2019, 50: 101657

[23] ZhaoH-M, ZhaoX, HanF-L, et al.. Cobalt crust recognition based on kernel Fisher discriminant analysis and genetic algorithm in reverberation environment [J]. Journal of Central South University, 2021, 28(1): 179-193

[24] Gayathri DeviK S, Sujatha ThereseP. Optimized PI controller for 7-level inverter to aid grid interactive RES controller [J]. Journal of Central South University, 2021, 28(1): 153-167

[25] SongJ-J, WangJ-Z, LuH-Y. A novel combined model based on advanced optimization algorithm for short-term wind speed forecasting [J]. Applied Energy, 2018, 215643-658

[26] LiangW-Z, LuoS-Z, ZhaoG-Y, et al.. Predicting hard rock pillar stability using GBDT, XGBoost, and LightGBM algorithms [J]. Mathematics, 2020, 8(5): 765

[27] LiL-J, YuY, BaiS-S, et al.. Towards effective network intrusion detection: A hybrid model integrating gini index and GBDT with PSO [J]. Journal of Sensors, 2018, 2018: 1578314

[28] MaoX-L, LiF-F, LiuX-Y, et al.. Detection of artificial pornographic pictures based on multiple features and tree mode [J]. Journal of Central South University, 2018, 2571651-1664

[29] ZhangX, WangX-R, ChenW, et al.A taxi gap prediction method via double ensemble gradient boosting decision tree [C], 2017, Beijing, China, IEEE, 255-260

[30] DragomiretskiyK, ZossoD. Variational mode decomposition [J]. IEEE Transactions on Signal Processing, 2014, 62(3): 531-544

[31] ZhangD, PengX-G, PanK-D, et al.. A novel wind speed forecasting based on hybrid decomposition and online sequential outlier robust extreme learning machine [J]. Energy Conversion and Management, 2019, 180: 338-357

[32] ThurowK, ChenC, JungingerS, et al.. Transportation robot battery power forecasting based on bidirectional deep-learning method [J]. Transportation Safety and Environment, 2020, 1(3): 205-211

[33] JaegerHThe “echo state” approach to analysing and training recurrent neural networks-with an erratum note [R], 2001, Bonn, Ger Ger Natl Res Cent Inf Technol, 148

[34] ZhongS-S, XieX-L, LinL, et al.. Genetic algorithm optimized double-reservoir echo state network for multi-regime time series prediction [J]. Neurocomputing, 2017, 238: 191-204

[35] LiuH, LongZ-H, DuanZ, et al.. A new model using multiple feature clustering and neural networks for forecasting hourly PM2.5 concentrations, and its applications in China [J]. Engineering, 2020, 6(8): 944-956

[36] LeaC, FlynnM D, VidalR, et al.Temporal convolutional networks for action segmentation and detection [C], 2017, Honolulu, HI, USA, IEEE, 10031012

[37] BAI Shao-jie, KOLTER J Z, KOLTUN V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling [OL]. arXiv preprint [2018-04-19]. https://arxiv.org/abs/1803.01271.

[38] HuJ-J, ZhouH, ZhouY-H, et al.. Flexibility prediction of aggregated electric vehicles and domestic hot water systems in smart grids [J]. Engineering, 2021, 7(8): 1101-1114

[39] HeY-D, ZhaoJ-B. Temporal convolutional networks for anomaly detection in time series [J]. Journal of Physics: Conference Series, 2019, 1213(4): 042050

[40] FriedmanJ H. Greedy function approximation: A gradient boosting machine [J]. The Annals of Statistics, 2001, 2951189-1232

[41] FriedmanJ H. Stochastic gradient boosting [J]. Computational Statistics & Data Analysis, 2002, 38(4): 367-378

[42] LiuH, DuanZ, WuH-P, et al.. Wind speed forecasting models based on data decomposition, feature selection and group method of data handling network [J]. Measurement, 2019, 148106971

[43] WuH-P, LiuH, DuanZ. PM_2.5 concentrations forecasting using a new multi-objective feature selection and ensemble framework [J]. Atmospheric Pollution Research, 2020, 11(7): 1187-1198

[44] LiY-F, LiuZ-Y, LiuH. A novel ensemble reinforcement learning gated unit model for daily PM_2.5 forecasting [J]. Air Quality, Atmosphere & Health, 2021, 14(3): 443-453

[45] SunW, LiZ-Q. Hourly PM_2.5 concentration forecasting based on mode decomposition-recombination technique and ensemble learning approach in severe haze episodes of China [J]. Journal of Cleaner Production, 2020, 263121442