Table 2

Results comparisons between DSTG and other methods, where ST indicates spatio-temporal.

Type Methods Scores MAE RMSE MAPE
S / T $ \mathcal{S}/\mathcal{T}$ GRU[tw] 42.4213 37.2985 47.5441 14.21%
AutoFormer[tw] 48.4166 42.8654 53.9678 27.40%
SCINet[tw] 42.1947 38.0369 46.3524 17.68%
Mic-LSTM[tw] 43.4459 39.2348 47.6571 13.64%
Mvmd-DNN[tw] 41.9165 37.5376 46.2957 9.45%
S / T $ \mathcal{S}/\mathcal{T}$-CNN GWNET[tw] 49.3725 43.1823 55.5627 14.12%
Ast-GCN[tw] 44.3207 39.8564 48.7850 10.35%
RBCNN[tw] 48.6482 43.2697 54.0266 13.22%
HSTTN[tw] 43.7319 38.3510 49.1128 11.76%
S / T $ \mathcal{S}/\mathcal{T}$-RNN DCRNN[tw] 44.1880 39.8136 48.5623 9.18%
AGCRN[tw] 46.7399 41.0236 52.4561 10.22%
GCMCN[tw] 43.3202 39.7753 46.8650 10.63%
DSTG[tw] 41.1007 36.6652 45.5362 8.40%
*

T $ \mathcal{T}$ represents the temporal aspect, while S $ \mathcal{S}$ denotes the spatial aspect.

**

The best metrics are bolded and those with † Indicate significant differences.

***

Methods use external temperature, wind speed or both as inputs are labeled [t], [w] or [tw], respectively.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.