In this paper, we proposed a method for optimising the deep long short term memory (LSTM) model to improve the quality of PM2.5 concentration prediction and used it for PM2.5 concentration prediction.
The main conclusions were drawn as follows:
In order to design a deep LSTM model optimised for PM2.5 concentration prediction, an index suitable for PM2.5 concentration prediction was selected and the number of layers was determined by an experimental method.
Also, the deep LSTM model was designed by determining the optimised number of units through genetic algorithm, and then applied to the PM2.5 prediction, resulting in the RMSE value of 2.341%, which validated its performance better than the non-optimised deep LSTM..
Finally, optimised deep LSTM model was compared with RNN and GRU models, which resulted in RMSE values in the PM2.5 concentration prediction of 2.341%, 2.568%, and 2.442%, respectively.
This study is used not only for PM2.5 concentration prediction, but also in various other fields and it is a valuable result for researchers who study environmental prediction and forecast.
Our result was published in the journal "Int. J. Environment and Pollution" under the title of "An improvement of PM2.5 concentration prediction using optimised deep LSTM"(http://doi.org/10.1504/IJEP.2021.10051956).
