Direct Augmented Lagrangian Evolution Strategies
Jeremy Porter, Dirk V. Arnold · Proceedings of the Genetic and Evolutionary Computation Conference · 2024
Existing evolutionary and swarm based algorithms that employ augmented Lagrangian techniques to solve constrained black-box optimization problems update the Lagrange multipliers using either tailored heuristics or variations of the prescription underlying the method of multipliers. We introduce an evolution strategy relying on an exact Lagrangian update instead, determining multiplier values and penalties from population based estimates of derivative related values. Numerical experiments on multiple test problems illustrate the potential of the approach.