Convolutional Recurrent Neural Network for Fault Diagnosis of High-Speed Train Bogie

Table 4

Hyperparameters used for comparison experiments.

(a) The architectures based on CNN, RNN, and CRNN

	HST SDS200
	CRNN	1D-CNN	LSTM
Conv_1d cells	116/232	116/232	-
pool_1d size	3	3	-
RNN cells	450/450/450/450	-	450/450/450/450
1D-CNN layers	2	2	-
RNN layers	4	-	4
Fully-connected layers	1	1	1
learning rate	0.005	0.0003	0.001

(b) The architectures based on ensemble learning

	RF	XGboost	GBDT
estimators size	1000	1000	1000
learning rate	-	0.15	0.01
max depth	500	3	8
max features	sqrt	-	sqrt