Prediction of air pollutant concentration based on self-attention mechanism LSTM model

In recent years, traditional deep learning has been widely used in the time series prediction of air quality, but this kind of model has many shortcomings in the input selection of meteorological related data. Based on the effectiveness of meteorological input data, this paper selects the daily meteorological data of 35 monitoring stations in Guiyang from 2019 to 2020. Analyzes the correlation between environmental data and six pollutant concentrations through Spearman Rank Correlation Coefficient; the ventilation factors and stable energy proposed from the perspective of energetics are added to the input elements to predict the concentration of pollutants. The experimental results show that the Long Short-term Memory network (SA-LSTM) model based on self attention mechanism with ventilation factors and stable energy is better than other existing models in air quality prediction.