Research Article
DWNet: Dual-Window Deep Neural Network for Time Series Prediction
Table 1
RMSE and MAE performance comparison among different methods and datasets (best result is displayed in boldface).
| Models | SML 2010 | Human Sports | EEG | Energy | RMSE | MAE | RMSE | MAE | RMSE | MAE | RMSE | MAE |
| ARIMA (16) | 0.2786 | 0.2219 | 0.1371 | 0.0617 | 0.5694 | 0.4724 | 0.8640 | 0.5740 | LSTM (16) | 0.1905 | 0.1489 | 0.0831 | 0.0325 | 0.2244 | 0.1724 | 0.6907 | 0.3663 | LSTM (128) | 0.2099 | 0.1671 | 0.0983 | 0.0437 | 0.3033 | 0.2283 | 0.8017 | 0.4376 | Encoder-Decoder (16) | 0.1438 | 0.0907 | 0.0774 | 0.0296 | 0.2499 | 0.1401 | 0.5983 | 0.2839 | Encoder-Decoder (128) | 0.1648 | 0.1012 | 0.0831 | 0.0303 | 0.4650 | 0.3036 | 0.6524 | 0.3117 | Input-Attn-RNN (16) | 0.1296 | 0.0762 | 0.0680 | 0.0282 | 0.2055 | 0.1447 | 0.5452 | 0.2564 | Input-Attn-RNN (128) | 0.1008 | 0.0897 | 0.0766 | 0.0362 | 0.4217 | 0.2881 | 0.5782 | 0.2502 | Temp-Attn-RNN (16) | 0.1097 | 0.0692 | 0.0646 | 0.0311 | 0.2220 | 0.1500 | 0.5414 | 0.2507 | Temp-Attn-RNN (128) | 0.1105 | 0.0770 | 0.0740 | 0.0334 | 0.3943 | 0.2998 | 0.5488 | 0.2563 | TCN (16) | 0.1156 | 0.0817 | 0.0628 | 0.0270 | 1.1845 | 0.9545 | 0.8279 | 0.5186 | TCN (128) | 0.1473 | 0.1136 | 0.0727 | 0.0329 | 1.1050 | 0.8696 | 0.8126 | 0.4567 | LSTNet (16) | 0.1277 | 0.0957 | 0.0582 | 0.0269 | 0.2322 | 0.1807 | 0.5733 | 0.2762 | LSTNet (128) | 0.1352 | 0.1020 | 0.0642 | 0.0312 | 0.2384 | 0.1868 | 0.6078 | 0.3296 | DARNN (16) | 0.0977 | 0.0644 | 0.0643 | 0.0232 | 0.1804 | 0.1442 | 0.5270 | 0.2439 | DARNN (128) | 0.1093 | 0.0778 | 0.0733 | 0.0435 | 0.3483 | 0.3250 | 0.5556 | 0.2525 | DSTP (16) | 0.0932 | 0.0614 | 0.0641 | 0.0227 | 0.1805 | 0.1414 | 0.5320 | 0.2459 | DSTP (128) | 0.0954 | 0.0670 | 0.0641 | 0.0235 | 0.1754 | 0.1384 | 0.5456 | 0.2525 | DWNet | 0.0891 | 0.0565 | 0.0575 | 0.0217 | 0.1702 | 0.1371 | 0.5015 | 0.2362 |
|
|
The window size of baseline methods is set to 16 and 128, and the short window size and long window size of DWNet are set to 16 and 128, respectively.
|