Gaussian process and interval gradient-centric evolutionary algorithm for dynamic multi-objective optimization

Dynamic multi-objective optimization problems (DMOPs) involve optimizing multiple conflicting objectives in environments where parameters or objectives vary over time. A primary challenge in addressing DMOPs is maintaining a balance between population diversity and convergence to the dynamic Pareto optimal front (POF). Traditional methods often struggle to adapt to environmental changes, which can result in significant performance degradation. To address this issue, we propose a novel evolutionary algorithm, termed the Gaussian process and interval gradient-centric evolutionary algorithm (GPIGC-EA). This algorithm integrates two key sub-strategies. The first, a GPR-based diversity preservation strategy (GPRDP), leverages Gaussian process regression (GPR) to predict the population’s boundary points. This prediction defines the range for generating new random individuals, thereby effectively enhancing population diversity. The second, an interval-based local centroid gradient prediction strategy (ILCGP), combines local gradient prediction with global center point prediction to guide the population’s convergence. By synergizing these strategies, the proposed framework effectively enhances population diversity while simultaneously capturing both global trends and local environmental dynamics. By incorporating both linear and nonlinear predictions, the framework ensures robust adaptation to dynamic environments. Experimental results demonstrate that GPIGC-EA outperforms or matches several state-of-the-art and classic dynamic multi-objective evolutionary algorithms on a majority of benchmark problems. This showcases its excellent stability and adaptability in addressing complex environmental changes.