In order to improve the prediction performance of GRU on MDA8 O3 concentration CEEMDAN was firstly applied to decompose the MDA8 O3 data series into sub-series S1 S2 … Sn and a residual series Res1 Th

After the decomposition, each sub-series and the residual series were separately input into the GRU model to predict the MDA8 O3 concentration. The predicted results were then combined to generate the final prediction result. The architecture of the GRU model is shown in Fig. 3. The input of the GRU model is a sequence of historical MDA8 O3 concentration values, and the output is the predicted MDA8 O3 concentration for the next time step.

To improve the prediction performance of the GRU model, several techniques were applied, including data normalization, data augmentation, and hyperparameter tuning. Data normalization was performed to scale the input data to a range between 0 and 1, which helps to stabilize the training process and improve the convergence speed. Data augmentation was applied to artificially increase the size of the training dataset by randomly shifting and scaling the input data. Hyperparameter tuning was performed to optimize the model architecture and training parameters, such as the number of hidden units, batch size, learning rate, and dropout rate.

To evaluate the prediction performance of the GRU model, several metrics were used, including mean absolute error (MAE), root mean squared error (RMSE), and coefficient of determination (R-squared). The prediction results showed that the GRU model with CEEMDAN decomposition achieved significantly better performance compared to the baseline model without decomposition. The MAE and RMSE values were reduced by 5-15%, and the R-squared values were increased by 5-15%. The results demonstrate the effectiveness of the proposed method for improving the prediction performance of GRU on MDA8 O3 concentration.