Generative Adversarial Network-based Missing Data Handling and Remaining Useful Life Estimation for Smart Train Control and Monitoring Systems

Table 7

Four RUL prediction methods.

RUL prediction method	GAN-based RUL estimation (the proposed method)	ARIMA-based RUL estimation	RUL estimation using “missing value pruning”	RUL estimation with “mean-value estimation” of missing values
Missing value handling mechanism	O (GAN-based data generation)	X (removal of records with missing values)	X (removal of records with missing values)	O (mean-value estimation)
RUL estimation method	Classification using GAN	ARIMA-based RUL estimation	Deep neural network	Deep neural network
Detailed parameters	Refer to Table 8	ARIMA (6, 2, 5)	(i) Learning epoch: 1000; (ii) Learning rate: (iii) Number of layers:10; (iv) Used activation functions; =(leaky RU for final layer, Sigmoid for layers 1−9)	—