Hybrid Time-Series Prediction Method Based on Entropy Fusion Feature

Table 3

Experiment settings of each control model.

Method	Hyperparameter setting
EEMD-ELSTM; EEMD-LSTM; ELSTM; LSTM	Hidden layer: 1; Hidden layer neuron: 40, 50; Learning rate: 0.0001; Batch size: 25, 5
Prophet	Changepoint prior scale: 64
	Interval width: 3
	Growth: linear
	Changepoints: 25
XGBoost	Estimators: 50
	Max depth: 6
	Min child weight: 0.5
	Eta: 0.2
ARIMA	P-autoregressive term: 2
	Q-moving average term: 2
	D-differential times: 0
SVR	Kernel: RBF
	Degree: 3
	C: 100
	Gamma: 1
LSTM-TCN	Filters: 128
	Kernel size: 3
	Hidden layer: 1
	Learning rate: 0.002
	Dilation rate: 1, 2, 4
	LSTM units: 64, 128, 256