Research Article
A Universal Concept for Robust Solving of Shortest Path Problems in Dynamically Reconfigurable Graphs
Table 5
Critical comparison of DVHNN and NAOP based concepts for SP determination.
| Comparison criteria | DVHNN based SP determination | NAOP based SP determination |
| External reconfigurability regarding dynamic change of edges’ weights | Yes | Yes |
| External reconfigurability regarding dynamic change of pairs | No | Yes |
| External reconfigurability regarding dynamic change of network topology | No | In principle yes, although this particular aspect has not been considered in this paper. It will be addressed in a future paper |
| Ability to cope with negative edges’ weights | No | Yes |
| Ability to cope with nonlinear path’s weights | No | In principle yes, although this particular aspect has not been considered in this paper. It will be addressed in a future paper |
| Reliability of the convergence: this stands for valid and successful convergence; otherwise, it is a failure | No Many failure cases have been observed; this is especially the case when high values of weights are present | Yes |
| Computational speed | Basically good if compared to the other NN based SP determination concepts | Very good (a difference of 1 to 2 order of magnitude better than DVHNN has been observed here) |
| Memory consumption need | High Because of the retraining need for each pair | Very low |
|
|
In this paper, the concepts have been all implemented in Matlab on a standard PC.
|