Multi-strategy attraction-repulsion optimization algorithm for engineering-oriented global optimization

The Attraction-Repulsion Optimization Algorithm (AROA) suffers from fixed attraction-repulsion strength and dimension-independent updates, which may cause dimensional isolation and an imbalance between exploration and exploitation. This article proposes a Multi-Strategy Attraction-Repulsion Optimization Algorithm (MSAROA). A Matrix Perturbation Competitive Differential Evolution (MPCDE) algorithm introduces cross-dimensional coupling through an adaptive transformation matrix and bidirectional crossover-competition selection. MPCDE is embedded within AROA to enable collaborative global and local search, while Elite Opposite-Based Learning improves population diversity and Levy flight helps escape local optima. Experiments on the CEC2017 (including CEC2020) and CEC2022 benchmark suites and six engineering optimization problems demonstrate that MSAROA achieves competitive convergence accuracy, speed and robustness. In particular, MSAROA obtains Friedman average ranks of 1.3161, 1.3103 and 1.5962 for the 30D, 50D and 100D CEC2017 problems, respectively, demonstrating its strong overall optimization performance.