Balancing Survival of Feasible and Infeasible Solutions in Evolutionary Optimization Algorithms

Handling constraints in any optimization algorithm is as important as an algorithm’s progress towards the optimal solution. Population-based optimization algorithms allow a flexible way to handle constraints by making a careful comparison of feasible and infeasible solutions present in the population. A previous approach, which emphasized feasible solutions infinitely more than the infeasible solutions, has been popularly applied for more than a decade, mostly with real-parameter genetic algorithms (RGAs). In this paper, we extended the approach in a way so as to make a balance between survival of feasible and infeasible solutions in a population. And the balancing was controlled through an additional parameter (α) that could be prespecified or adaptively updated as the algorithm progresses. The approach was incorporated in the RGA, a significant improvement in performance was observed in the g-series test problem suite and a real-world application problem (welded-beam). A further analysis on the effect of α was performed and some standard values of α are suggested based on our empirical findings.