Prediction of Air Pollution Using LSTM

Abstract

The paper describes the application of long short-term memory (LSTM) for air pollution forecasting. LSTM is a special design of deep structure recurrent neural networks, which is very well suited for the prediction of the sequences of data. This work investigates its properties in the task of the short-time one-hour ahead and the one day ahead prediction of air pollutants such as PM10, SO2, NO2, and ozone in Warsaw, Poland. The results of numerical investigations have shown very good accuracy in online prediction, exceeding the corresponding values obtained at the application of feedforward neural structures.