Table 3 Performance of only BPNN model without the DE analytic solution.
From: Concrete crack opening forecasting by back propagation neural network and differential equation
Item | JB-1 with mean −21.67 | JB-3 with mean −18.23 | JB-7 with mean 0.95 | ||||||
|---|---|---|---|---|---|---|---|---|---|
Train | Test | Forecast | Train | Test | Forecast | Train | Test | Forecast | |
The ratio < 20% error | 100% | 99.5% | 98.1% | 100% | 98.6% | 97.5% | 73.7% | 75.9% | 72.1% |
Average absolute error (% mm) | 0.013% | 0.014% | 0.014% | 0.07% | 0.075% | 0.079% | 14.99% | 14.57% | 14.68% |
Minimum error (% mm) | 0% | 0.001% | 0.003% | 0% | 0.02% | 0.06% | 0% | 0% | 0.02% |
Maximum error (% mm) | 0.060% | 0.059% | 0.073% | 0.219% | 0.354% | 0.329% | 85.45% | 85.45% | 84.86% |
R2 | 0.78 | 0.772 | 0.759 | 0.619 | 0.604 | 0.591 | 0.856 | 0.887 | 0.853 |
Kling Gupta efficiency | 0.838 | 0.834 | 0.816 | 0.700 | 0.695 | 0.698 | 0.897 | 0.892 | 0.885 |
