Adaptive Gene Expression Programming Algorithm Based on Cloud Model

Standard gene expression programming(GEP) works with fixed rate of mutation and crossover, ignoring the variation of the individual fitness, hence it works in the local optimum style with the low convergence speed. This paper aims to introduce cloud model to GEP. The main contributions include: (1) Formally describing the new concepts such as fitness degree, valid individual, the family measure and cloud mutation rate, etc. (2) Analysing mathematical properties for cloud mutation; (3) Proposing adaptive cloud strategy (ACS). It determines mutation and crossover rate dynamically; (4) Proposing valid crossover strategy (VCS) to keep good objects and improve the diversity; (5) Extensive experiments testify the better performance of the new method. The average fitness is increased by 9%, the minimal fitness is increased by 10% and the average generation for the best individual is decreased by 11%.