Unified Quantile Regression Deep Neural Network with Time-Cognition for Probabilistic Residential Load Forecasting

Table 5

Average efficiency comparison of 20 residences.

Item	QGBRT	QCNN	QLSTM	Proposed	R_QLSTM
Training time per epoch (s)	None	24	209	26	87.56%
Testing time (s)	0.18	0.88	16.27	0.89	94.53%
Training convergence epochs	None	77	68	66	—
Total training time (s)	1,517	1,814	14,212	1,716	87.93%
Parameters	None	184,681	330,889	187,769	—