Research Output
A Self-organizing LSTM-Based Approach to PM2.5 Forecast
  Nanjing has been listed as the one of the worst performers across China with respect to the high level of haze-fog, which impacts people's health greatly. For the severe condition of haze-fog, PM2.5 is the main cause element of haze-fog pollution in China. So it’s necessary to forecast PM2.5 concentration accurately. In this paper, an artificial intelligence method is employed to fore-cast PM2.5 in Nanjing. At the data pre-processing stage, the main factors among the air pollutants (O3, NO2, SO2, CO, etc.) as well as meteorological parameters (pressure, wind direction, temperature, etc.) that affect PM2.5 are selected, and these factors of previous hours are as input data to predict PM2.5 concentration of next hours. Considering the air pollutants and meteorological data are typical time series data, a special recurrent neural network, which is called long short term memory (LSTM) network, is applied in this paper. To determine the amount of nodes in the hidden layer, a self-organizing method is used to automatically adjust the hidden nodes during the training phase. Finally, the PM2.5 concentrations of the next 1 hour, 4 hours, 8 hours, and 12 hours are predicted separately by using the self-organizing LSTM network based approach. The experimental result has been validated and compared to other algorithms, which reflects the proposed method performs best.

  • Date:

    13 September 2018

  • Publication Status:


  • DOI:


  • Library of Congress:

    QA75 Electronic computers. Computer science

  • Dewey Decimal Classification:

    005 Computer programming, programs & data

  • Funders:

    European Commission


Liu, X., Liu, Q., Zou, Y., & Wang, G. (2018). A Self-organizing LSTM-Based Approach to PM2.5 Forecast. In Proceedings of the 4th International Conference on Cloud Computing and Security (ICCCS 2018), (683-693).



Haze-fog, PM2.5 Forecasting, Selecting Main Factors, Time Series Data, LSTM network, Self-organizing Algorithm

Monthly Views:

Linked Projects

Available Documents