The complex problem of wireless network planning can be tackled by specific software but this applications usually require high computation times and resources: in order to reduce this computation burden and to increase efficiency, the EML2 group has developed a multi-agent system named “Genetic Agents”.
In this system, advanced IT techniques and pre-existing software codes for the EM evaluation has been merged. Software agents, deployed in a computational grid parallelize a genetic algorithm in order to optimize results given by an EM simulator.
This particular genetic algorithm implementation is defined “Parallel Genetic Algorithm (PGA) island-based” because the original sequential code is replicated in each node (that means in each island) of the grid. Moreover, the initial population of the genetic algorithm is partitioned into subpopulations; each of them is distributed in a different island. This technique led to reduced computational times and to better explorations in the solution space. In general, we have a better management of the global search rather than in the sequential case.
As for the implementation: electrical and positioning parameters of the antennas in the wireless network that has to be analyzed and optimized have been codified in binary strings (chromosomes). Genetic operators are applied on the chromosomes in the typical way. However, the genetic material exchange between different islands (that led to a better exploration of the solution space) is performed by software agents.
Results (depicted in the figure below) show that this configuration has reduced computational times and the EM levels measured in a given area; in addition, a better quality of service in the network has been achieved.