Table 2.
Improved load classification rates by the parallel GA-embodied ANN, in this experiment, considering 2,505,000 feed-forward ANNs differing in their trainable weight coefficients.
|
Population Sizes (the maximum number of generations: 500) |
|||||
| 1000 (501,000 function evaluations) |
2000 (1,002,000 function evaluations) |
5000 (2,505,000 function evaluations) |
|||
| the configured ANN evolved by the standard GA | |||||
| Computation time 1 (mins) | 336.52 | 745.65 | 4043.23 | ||
| Averaged Sum of Squared Errors (SSE) | 0.40 | 0.40 | 0.47 | ||
| the configured ANN evolved by the parallel GA | |||||
| Computation time (mins) | 6.94 | 12.20 | 29.91 | ||
| Averaged SSE | 0.38 | 0.36 | 0.25 | ||
| Load classification rates by the configured ANN evolved by the different types of GA considering the different sizes of population (%) | Improvement in load classification rate (%) | ||||
| Electrical appliance |
the configured ANN evolved by the standard GA (population size: 1000) |
the configured ANN evolved by the parallel GA (population size: 5000) |
|||
| electric rice cooker | 74.06 | 83.75 | +9.69 | ||
| electric water boiler | 79.17 | 94.69 | +15.52 | ||
| steamer | 88.54 | 90.83 | +2.29 | ||
| TV | 89.69 | 98.96 | +9.27 | ||
1 The computation time increases exponentially. Meta-heuristically training/evolving the configured feed-forward ANN in a serial/sequential execution is extremely computationally intensive.