Prediction of North Atlantic Oscillation Index Associated with the Sea Level Pressure Using DWT-LSTM and DWT-ConvLSTM Networks

Figure 15

The sketch map of the rolling forecasting, which is applied in multistep prediction. Each piece of training data’s length is set to the lagged coefficient determining in Section 2.3. Suppose that we put the first piece of training data into the model and obtain the predicted value ; then the next input is generated by removing the first frame and appending to the end of input. Putting the next input into the model, we obtain a new predicted value . Repeating the above steps, the predicted sequence is .