Genetic Improvement of Dynamic Optimization Algorithms using PushGP

In this paper the idea of automatic improvement of existing heuristic optimization algorithms using genetic programming is investigated. In particular, the Push genetic programming system is used to generate new local search heuristics for the multi-swarm quantum swarm optimization method using the generalized moving peaks benchmark. The best designed heuristic is then applied to a more advanced dynamic optimizer, the adaptive multi-population particle swarm optimization algorithm. The results indicate that the generated heuristic, being unlike any known, allows significant improvement of algorithm performance on most of the test problems of the CEC 2025 competition on dynamic optimization generated by generalized moving peaks benchmark.