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).

ModelsSML 2010Human SportsEEGEnergy
RMSEMAERMSEMAERMSEMAERMSEMAE

ARIMA (16)0.27860.22190.13710.06170.56940.47240.86400.5740
LSTM (16)0.19050.14890.08310.03250.22440.17240.69070.3663
LSTM (128)0.20990.16710.09830.04370.30330.22830.80170.4376
Encoder-Decoder (16)0.14380.09070.07740.02960.24990.14010.59830.2839
Encoder-Decoder (128)0.16480.10120.08310.03030.46500.30360.65240.3117
Input-Attn-RNN (16)0.12960.07620.06800.02820.20550.14470.54520.2564
Input-Attn-RNN (128)0.10080.08970.07660.03620.42170.28810.57820.2502
Temp-Attn-RNN (16)0.10970.06920.06460.03110.22200.15000.54140.2507
Temp-Attn-RNN (128)0.11050.07700.07400.03340.39430.29980.54880.2563
TCN (16)0.11560.08170.06280.02701.18450.95450.82790.5186
TCN (128)0.14730.11360.07270.03291.10500.86960.81260.4567
LSTNet (16)0.12770.09570.05820.02690.23220.18070.57330.2762
LSTNet (128)0.13520.10200.06420.03120.23840.18680.60780.3296
DARNN (16)0.09770.06440.06430.02320.18040.14420.52700.2439
DARNN (128)0.10930.07780.07330.04350.34830.32500.55560.2525
DSTP (16)0.09320.06140.06410.02270.18050.14140.53200.2459
DSTP (128)0.09540.06700.06410.02350.17540.13840.54560.2525
DWNet0.08910.05650.05750.02170.17020.13710.50150.2362

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.