*4.2. Division of Dataset*

To train the prediction model parameters, which are mainly some structural weight values, 75% of historical data samples were recognized as the training dataset, and the remaining 25% of data samples were taken as the testing dataset to examine the prediction efficiency. The ensemble division of dataset is shown in Figure 5.

As shown in Figure 5, the training process adopts a cross validation mechanism, composed by many epochs. In each epoch, 90% of the training samples are regarded as a sub-training set, and the remaining 10% of the training samples are regarded as the subtesting dataset. The partition scheme of the sub-training dataset and sub-testing dataset is to divide them randomly. From Figure 5, it can be found that the optimal parameters are obtained through multiple cross-validation, which was used to provide a basis for the subsequent experiments.

**Figure 5.** Division of the dataset.
