Table 3 Training times in hours, measured on a machine with a Nvidia 3090 RTX GPU
From: Transformer training strategies for forecasting multiple load time series
Model | Electricity | Ausgrid | ||||
|---|---|---|---|---|---|---|
24h | 96h | 720h | 24h | 96h | 720h | |
Linear regression (local) | 0.02 | 0.03 | 0.08 | 0.02 | 0.03 | 0.07 |
MLP (local) | 0.42 | 0.42 | 0.36 | 0.40 | 0.39 | 0.39 |
LSTM (multivariate) | 0.06 | 0.08 | 0.30 | 0.03 | 0.04 | 0.10 |
LSTM (local) | 8.25 | 7.61 | 7.20 | 4.27 | 3.55 | 3.49 |
LSTM (global) | 1.09 | 0.82 | 0.98 | 1.11 | 0.84 | 0.71 |
Transformer (multivariate) | 0.19 | 0.23 | 0.88 | 0.10 | 0.09 | 0.39 |
Transformer (local) | 14.20 | 16.82 | 102.09 | 8.33 | 9.74 | 62.53 |
Transformer (global) | 3.42 | 2.00 | 9.86 | 3.85 | 2.85 | 6.77 |