Evolutionary Optimization of High-Dimensional Multiobjective and Many-Objective Expensive Problems Assisted by a Dropout Neural Network

Gaussian processes (GPs) are widely used in surrogate-assisted evolutionary optimization of expensive problems mainly due to the ability provide a confidence level their outputs, making it possible adopt principled surrogate management methods, such as acquisition function Bayesian optimization. Unfortunately, GPs become less practical for high-dimensional multiobjective and many-objective computational complexity is cubic number training samples. In this article, we propose computationally efficient dropout neural network (EDN) replace process new model strategy achieve good balance between convergence diversity assisting algorithms solve problems. While conventional needs save large models during calculating level, only one single needed EDN estimate fitness its by randomly ignoring neurons both testing network. Extensive experimental studies on benchmark with up 100 decision variables 20 objectives demonstrate that, compared state art, proposed algorithm not highly competitive performance but also more scalable Finally, validated an operational problem crude oil distillation units, further confirming capability handling given limited budget.