Fig 1.
Traditional load forecasting model based on multi-dimensional time series data, where historical features are used to predict future load values.
Fig 2.
Time-series Load Graph Model, where node features represent load values and edge features represent other related variables.
Fig 3.
Training framework of the EGAT-based load forecasting model, including graph-based feature extraction and load prediction.
Table 1.
Case 1 simulation results.
Fig 4.
Single-step load forecasting performance comparison of different models in Case 1.
Fig 5.
Fitting curves of predicted and actual load values on the test dataset.
Table 2.
Case 2 simulation results.
Fig 6.
Multi-step load forecasting performance comparison of different models in Case 2.
Fig 7.
Load forecasting fitting curves for four typical days representing different seasons.
Table 3.
Case 3 simulation results.
Fig 8.
Single-step load forecasting performance comparison of different models in Case 3.
Table 4.
Case 1 ablation study results.
Fig 9.
Single-step load forecasting performance comparison of different models in the ablation study (Case 1).
Table 5.
Case 2 ablation study results.
Fig 10.
Muti-step load forecasting performance comparison of different models in the ablation study (Case 2).
Table 6.
Case 3 ablation study results.
Fig 11.
Single-step load forecasting performance comparison of different models in the ablation study (Case 3).