Novel Hybrid Algorithm Integrating BAS and PSO for Enhanced Optimization

Beetle Antennae Search (BAS) and Particle Swarm Optimization (PSO) represent two classical intelligent optimization algorithms. Their integration leverages complementary strengths to enhance optimization performance. BAS mimics the antennae sensing mechanism of beetles, employing stochastic search to approximate optimal solutions with advantages including fast convergence and minimal parameter requirements. PSO simulates bird flock foraging behavior, achieving global optimization through information sharing among particles. In code implementation, BAS typically requires only 2-3 parameters (step size and sensing range) while PSO maintains velocity and position updates for each particle in the swarm.

The BAS-PSO hybrid addresses limitations of individual algorithms. BAS demonstrates excellent local search capabilities but tends to fall into local optima, whereas PSO excels at global exploration but suffers from slower convergence. The novel algorithm incorporates BAS's local refinement strategy within PSO's global search framework. Implementation-wise, this involves modifying the PSO update rules to include BAS-inspired local search steps when particles approach promising regions, maintaining global exploration while accelerating convergence to optimal solutions. The hybrid approach typically uses PSO for initial global exploration and switches to BAS-style fine-tuning when convergence indicators are met.

Simulation experiments confirm the algorithm's applicability across diverse optimization models including function optimization, neural network training, and parameter tuning. Comparative studies validate its enhanced performance in high-dimensional complex problems, particularly improvements in convergence accuracy and computational efficiency. Code implementation often features adaptive switching mechanisms between global and local search phases. Future research directions may include adaptive parameter adjustment, multi-strategy fusion, and applications to large-scale optimization problems, potentially implemented through dynamic parameter matrices and parallel computing techniques.