A Hybrid Model for Air Quality Prediction Based on Data Decomposition

Accurate and reliable air quality predictions are critical to the ecological environment public health. For traditional model fails make full use of high low frequency information obtained after wavelet decomposition, which easily leads poor prediction performance model. This paper proposes a hybrid based on data choosing decomposition (WD) generate high-frequency detail sequences WD(D) low-frequency approximate WD(A), using sliding window for reconstruction processing, long short-term memory (LSTM) neural network autoregressive moving average (ARMA) WD(A) prediction. The final results can be by accumulating predicted values each sub-sequence, reduces root mean square error (RMSE) 52%, absolute (MAE) 47%, increases goodness fit (R2) 18% compared with single Compared mixed model, reduced RMSE 3%, MAE increased R2 0.5%. experimental verification found that proposed solves problem lagging is feasible method.