Time Series Prediction of AQI in Wuhan City Using a Hybrid Prophet-LSTM Model with an Improved PSO Algorithm

The air quality index (AQI) depends on the concentrations of six pollutants (PM_2.5, PM₁₀, SO₂, NO₂, O₃, and CO). In this paper, a Prophet-LSTM model with improved particle swarm optimization (PSO) is proposed to analyze the time series of six pollutant concentrations in Wuhan city. First, the time series are decomposed by Prophet, and Prophet is used to predict the trend term and periodic term. Then, LSTM is used to predict the error term. Finally, the improved PSO algorithm is used for optimization. These experimental results indicated that (1) Prophet’s decomposition method has good applicability to time series with the multiplication form. The Prophet-LSTM model can overcome the influence of PM series irregularity, large fluctuations and multiple noise on the prediction effect, which improves the prediction ability of the model. (2) The improved PSO algorithm can greatly improve the accuracy of the weight solution space and has the attribute of parallel computing, which makes the solution forms more diversified. (3) The hybrid model has better prediction ability than the comparison model (LSTM, Prophet, Prophet-LSTM). The hybrid model combines the advantages of Prophet and LSTM, which has strong adaptability to the randomness of sample selection and has strong accuracy in predicting pollutant concentrations.