Table 5
Performance comparison of 15 ANN with different BLPA and number of neurons in the hidden layer.
| BPLA | Function | Na | Epochs | MSEb | (R2)c |
|---|---|---|---|---|---|
| BFGS quasi-Newton back-propagation | Trainbfg | 5 | 9 | 4.4748 × 10−5 | 0.8704 |
| Conjugate gradient back-propagation with Powell-Beale restarts | Traincgb | 5 | 31 | 7.6523 × 10−6 | 0.9934 |
| Conjugate gradient back-propagation with Fletcher-Reeves updates | Traincgf | 5 | 32 | 3.5988 × 10−5 | 0.9823 |
| Conjugate gradient back-propagation with Polak-Ribiére updates | Traincgp | 5 | 17 | 1.1361 × 10−4 | 0.9327 |
| Gradient descent with adaptive learning rate back-propagation | Traingda | 5 | 240 | 7.4223 × 10−5 | 0.9233 |
| Gradient descent with momentum back-propagation | Traingdm | 5 | 1000 | 1.2444 × 10−4 | 0.4658 |
| Gradient descent with momentum and adaptive learning rate back-propagation | Traingdx | 5 | 261 | 7.8924 × 10−5 | 0.9414 |
| Levenberg-Marquardt back-propagation | Trainlm | 5 | 10 | 2.4115 × 10−6 | 0.9953 |
| One-step secant back-propagation | Trainoss | 5 | 9 | 1.0892 × 10−4 | 0.8508 |
| Resilient back-propagation | Trainrp | 5 | 40 | 2.1994 × 10−5 | 0.9922 |
| Scaled conjugate gradient back-propagation | Trainscg | 5 | 15 | 4.9185 × 10−5 | 0.9517 |
| Levenberg-Marquardt back-propagation | Trainlm | 2 | 24 | 9.6908 × 10−5 | 0.9833 |
| Levenberg-Marquardt back-propagation | Trainlm | 3 | 17 | 9.6355 × 10−6 | 0.9936 |
| Levenberg-Marquardt back-propagation | Trainlm | 4 | 16 | 9.6565 × 10−6 | 0.9919 |
| Levenberg-Marquardt back-propagation | Trainlm | 6 | 11 | 1.0432 × 10−5 | 0.9923 |