Representation matters: Real-valued encodings for meander line RFID antennas

Abstract

Solution representation can have a large impact on the performance of heuristic solvers. When tackling bounded self-avoiding walk problems, such as the meander line RFID antenna design problem, solutions may be represented in terms of the absolute or relative direction of travel at each step. Encoding these instructions in a continuous space is required in order to apply continuous solvers, but also allows for an adaptive interpretation of each instruction that promotes longer paths. Using path length as a proxy for antenna quality, this work demonstrates that the adaptive solution representations outperform their non-adaptive counterparts, and that starting from a corner node in the square design space positively influences algorithm performance. The superior performance of a relative encoding over an absolute one is confirmed, in both the single objective of maximising path length and in a substantial investigation of the multi-objective antenna design problem. In the multi-objective case, simplifying the problem by fixing the antenna start node can assist the algorithm to perform well, but allowing the algorithm to evolve antennas starting from any point leads to more consistent results.