Research Article

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

Table 4

AQS comparison of three time-coding methods.

ResidenceNature codingOne-hot codingPeriodic codingI_nature (%)I_one-hot (%)

10.06040.05920.05469.677.83
20.12610.12260.12064.401.67
30.14010.14490.13920.703.94
40.18230.15850.153615.753.10
50.07780.06660.065116.252.15
60.08850.09510.08563.299.99
70.14130.13780.13087.495.12
80.23790.21140.210111.660.60
90.08880.08950.08118.689.42
100.10910.10180.09839.913.47
110.05230.04750.046111.892.91
120.11760.11160.10788.343.38
130.04330.04010.038411.274.21
140.09830.09460.09018.264.67
150.09330.08960.080913.279.75
160.11880.12390.11404.038.00
170.02590.02530.022612.8910.63
180.07070.06300.063210.63āˆ’0.30
190.12000.12040.10919.099.34
200.03510.03490.03199.018.51
Average0.10140.09690.09229.104.91