Combined Prediction Model of PM2.5 Concentration Based on Wavelet Transform and LSTM

Abstract

To improve the prediction accuracy of PM2.5 concentration and overcome the applicability limitation of a single prediction method, a combined prediction model of PM2.5 concentration based on wavelet transform and long short-term memory (LSTM) is proposed. The potential relationship between PM2.5 and other air quality indexes is explored by using feature engineering, and strong correlation factors are selected as input vectors. Wavelet transform is used to decompose and refine the time series of related factors at multiple scales. And the LSTM is used to train the time series in different scales, and the final prediction results are generated through reconstruction. The results show that the combined model has a better prediction effect compared with the single prediction model, which improves the prediction accuracy and model generalization ability.