Choosing neural network models for short-term quality forecasting of atmospheric air with a limited amount of data

A.A. Egorkin1,2,3

1Institute of Natural and Technical Systems, RF, Sevastopol, Lenin St., 28

2FGAOU VO “Sevastopol State University”, RF, Sevastopol, Universitetskaya St., 33

3FGKVOU VPO “Black Sea Higher Naval Order of the Red Star School named after P.S. Nakhimov”,

 RF, Sevastopol, Dybenko St., 1a

DOI: 10.33075/2220-5861-2023-4-123-134

UDC 504:004.89

EDN: https://elibrary.ru/vvpogx

Accurate forecasts of environmental pollution and, in particular, atmospheric air are necessary for making sound management decisions to reduce this negative impact. Currently, the use of monitoring systems based on inexpensive sensor devices in their design is becoming widespread. This allows you to build a more saturated spatial monitoring network with an acceptable cost. But real-time data is not enough to predict the environmental situation. It is promising to make a forecast using machine learning methods and monitoring data together. The work is aimed at a comprehensive study of the possibility of short-term forecasting of concentrations of pollutants (particulate matter) with a limited amount of data using modern machine learning (ML) methods. Machine learning models such as TFT, LSTM, RNN, GPU, TiDE, KAN, TCN, BiTCN, NHits and NBeats are investigated to consider the possibility of obtaining an acceptable forecast corresponding to the selected quality indicators. The study shows that the NBeats, TCN and NHits models demonstrate the potential for use in predicting pollutant concentrations, given the limited amount of data. The use of the NBeats, TCN and NHits models, as the most acceptable in terms of quality, can be recommended for use in decision support systems after working through the selection of hyperparameters of the models.

Keywords: particulate matter, atmospheric pollution, atmospheric air monitoring systems, atmospheric air condition forecast, neural network technologies.

Full text in PDF(RUS)

REFERENCES

  1. Montalvo L., Fosca D., Paredes D., Abarca M., Saito C., and Villanueva E. An air quality monitoring and forecasting system for Lima city with low-cost sensors and artificial intelligence models. Frontiers in Sustainable Cities, 2022, Vol. 4. DOI: 10.3389/frsc.2022.849762.
  2. Masood A. and Ahmad K. A model for particulate matter (PM2.5) prediction for Delhi based on machine learning approaches. Procedia Computer Science, 2020, Vol. 167 (2), pp. 2101–2110.
  3. https://github.com/Nixtla/neuralforecast.
  4. Vaswani A., Shazeer N., Parmar N., Uszkoreit J., Jones L., Gomez A.N., Kaiser L. and Polosukhin I. Attention is all you need. Neural Information Processing Systems, 12 June 2017.
  5. Elman J.L. Finding structure in time. Cognitive Science, 1990, Vol. 14 (2), pp. 179–211.
  6. Hochreiter S. and Schmidhuber J. Long short-term memory. Neural Computation, 1997, Vol. 9 (8), pp. 1735–1780.
  7. Bolis S., Guerini M., and Pasta A. Nhits: A neural hits model for sales fore-casts in e-commerce. 2021.
  8. Oreshkin B.N., Carpov D., Chapados N., and Bengio Y. N-beats: Neural basis expansion analysis for interpretable time series forecasting. Proceedings of International Conference on Learning Repre-sentations, 2020, Preprint.
  9. Das A., Kong W., Leach A., Sen R., and Yu. R. Long-term Forecasting with TiDE: Time-series Dense Encoder, 2023, Preprint. DOI:10.48550/arXiv.2304.08424.
  10. Bai S., J. Zico Kolter, and Koltun V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. Computing Research Repository. Proceedings of the 1st Inter-national Workshop on Multimedia Content Analysis in Sports, 2018, pp. 11–19. Available at: https://arxiv.org/abs/1803.01271.
  11. Cho K. On the Properties of Neural Machine Translation: Encoder-decoder Approaches. ACL Anthology, 2014, pp. 103–111.
  12. Liu Z. et al. Kan: Kolmogorov-Arnold networks. Available at: https://doi.org/10.48550/arXiv.2404.19756
  13. Sprangers O., Schelter S., and de Rijke M. Parameter-efficient deep probabilistic forecasting. International Journal of Forecasting, 2023, Vol. 39 (1), pp. 332–345.

Loading