Inverse Gaussian Process Modeling for Evolutionary Dynamic Multiobjective Optimization.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Machine learning

Background:

• Dynamic multiobjective optimization problems (DMOPs) present challenges in tracking evolving Pareto-optimal fronts.
• Traditional methods often estimate solutions in the decision space, which may not align with decision-maker preferences in the objective space.
• Existing inverse model-based algorithms struggle with nonlinear correlations in DMOPs, limiting their precision.

Purpose of the Study:

• To propose a novel inverse Gaussian process (IGP)-based prediction approach for solving DMOPs.
• To enhance the precision and applicability of inverse models for nonlinear DMOPs.
• To improve the diversity and convergence of solutions in the objective space for DMOPs.

Main Methods:

• An inverse Gaussian process (IGP) model is developed to map historical optimal solutions from the objective space to the decision space.
• A sampling mechanism generates sample points in the objective space.
• The IGP predictor is utilized to create an effective initial population using these sample points.

Main Results:

• The proposed IGP-based approach yields solutions with improved diversity and convergence in the objective space compared to traditional methods.
• The method demonstrates superior performance over existing inverse model-based techniques for nonlinear DMOPs.
• Experimental results on benchmark problems and a real-world raw ore allocation problem confirm significant improvements in dynamic optimization performance.

Conclusions:

• The inverse Gaussian process (IGP) approach offers a more responsive and effective solution for dynamic multiobjective optimization problems.
• This method addresses the limitations of traditional approaches and existing inverse models, particularly for nonlinear DMOPs.
• The proposed algorithm shows practical significance for real-world dynamic optimization challenges.