Study on an Improved Harris Hawks Optimization Algorithm for Control Parameter Tuning

To address the characteristics of control parameter tuning problems, such as narrow feasible regions, strong boundary sensitivity, significant objective-function nonlinearity, and high requirements for late-stage high-precision search, this paper proposes an improved Harris Hawks Optimization (HHO) algorithm for engineering-oriented tuning tasks. Built upon the standard HHO framework, the proposed method introduces three major improvements: initialization enhancement, stagnation recovery, and late-stage refinement. First, a hybrid initialization strategy that integrates global random sampling, prior-region guidance, and opposition-based learning is adopted to improve the coverage of the initial population over narrow feasible regions and boundary-adjacent high-quality areas. Second, a stagnation-awareness mechanism is constructed based on both the improvement magnitude of the best fitness and population diversity, and a directed escape strategy is incorporated to restore search activity in the middle stage of evolution. Finally, an elite-guided local refinement mechanism is introduced, where coordinate-diagonal hybrid search and random perturbation polishing are employed to enhance final solution accuracy and result stability. The proposed algorithm is validated on two representative control tuning problems, namely LLC and AVR, and its generalization capability is further examined on the supplementary THIRD and MMS scenarios. The results show that the improved HHO achieves better median performance and lower tail risk in both LLC and AVR cases. It is particularly suitable for control parameter tuning problems characterized by narrow feasible regions, pronounced boundary high-quality solutions, and strong demands for terminal refinement.