An adaptive response algorithm based on dual-space detection for dynamic multiobjective optimization

Efficiently tracking the dynamically changing Pareto-optimal set (POS) or Pareto-optimal front (POF) is a core task in dynamic multiobjective optimization. Most dynamic multi-objective evolutionary algorithms (DMOEAs) implement dedicated response mechanisms to mitigate the impact of environmental changes. To address the critical yet underexplored impact of varying change severities in both the POS and the POF, we propose an adaptive response algorithm based on dual-space detection, named ARA-DMOEA. Our approach incorporates a dual-space change severity detection mechanism that quantifies POS and POF variations, dynamically classifying changes as minor or significant. Based on this real-time assessment, ARA-DMOEA adaptively activates tailored response strategies. Specifically, when significant changes are detected in either space, a Gated Recurrent Unit (GRU) prediction model generates high-quality initial populations by leveraging historical solution patterns. For minor POS changes, a Centroid-guided Differential Prediction (CDP) strategy exploits population shift trends to maintain solution diversity. For minor POF changes, a Random Solution Generation (RSG) strategy enhances diversity by expanding sampling ranges around predicted ideal and nadir points. By synergistically combining these strategies according to dual-space severity detection, ARA-DMOEA dynamically optimizes its response to environmental shifts. In comparison with six state-of-the-art algorithms on a series of dynamic multiobjective problems, ARA-DMOEA demonstrates superior adaptability to environmental changes while achieving better convergence and diversity.