Table 1 The transfer function’s effect on the mode’s prediction accuracy.
From: Temperature prediction of solar greenhouse based on NARX regression neural network
Order number | Transfer function | Training step | RMS error | Coefficient of determination | Training time | |
|---|---|---|---|---|---|---|
Implication level | Output level | |||||
1 | radbas | radbas | 8 | 113.0576 | 0.7486 | 18.049 |
2 | radbas | logsig | 13 | 137.0880 | 0.1512 | 21.977 |
3 | radbas | tansig | 13 | 0.2326 | 0.9978 | 22.603 |
4 | radbas | purelin | 13 | 0.0224 | 0.9998 | 16.719 |
5 | logsig | radbas | 14 | 142.0744 | 0.06037 | 22.967 |
6 | logsig | logsig | 10 | 112.9237 | 0.7537 | 23.047 |
7 | logsig | tansig | 21 | 0.9222 | 0.991 | 20.620 |
8 | logsig | purelin | 7 | 0.0166 | 0.9998 | 18.409 |
9 | tansig | radbas | 7 | 116.1256 | 0.6604 | 22.727 |
10 | tansig | logsig | 25 | 112.8936 | 0.7565 | 24.207 |
11 | tansig | tansig | 19 | 0.0611 | 0.9995 | 23.625 |
12 | tansig | purelin | 15 | 0.0133 | 0.9999 | 19.9925 |
13 | purelin | radbas | 6 | 113.5387 | 0.7458 | 27.386 |
14 | purelin | logsig | 28 | 113.5672 | 0.7348 | 40.981 |
15 | purelin | tansig | 5 | 1.3032 | 0.9878 | 23.410 |
16 | purelin | purelin | 22 | 0.0172 | 0.9998 | 18.203 |
