Table 4.
Convolution LSTM neural network model statistics | ||||||
---|---|---|---|---|---|---|
Prediction size (forecast) | Train RMSE [°C] | Test RMSE [°C] | Train r2 | Test r2 | Train F1 | Test F1 |
6 h | 1.57 | 1.57 | 0.98 | 0.98 | 0.95 | 0.96 |
12 h | 1.83 | 1.88 | 0.98 | 0.98 | 0.95 | 0.95 |
24 h | 2.18 | 2.19 | 0.97 | 0.97 | 0.94 | 0.95 |
30 h | 2.42 | 2.66 | 0.96 | 0.95 | 0.94 | 0.93 |
36 h | 2.57 | 2.89 | 0.95 | 0.95 | 0.93 | 0.92 |
48 h | 2.82 | 3.06 | 0.95 | 0.94 | 0.93 | 0.93 |
The CNN model uses two convolution layers and one LSTM layer configuration.