Particle swarm optimization (PSO) is envisioned as potential solution to overcome maximum power point tracking (MPPT) problems. Nevertheless, conventional PSO suffers from large transient oscillation, slow convergence and tedious parameter tuning when tracking global MPP (GMPP) under partial shading conditions (PSC), leading to poor efficiency and significant power loss. Therefore, a modified PSO hybridized with adaptive local search (MPSO-HALS) is designed as a robust, real-time MPPT algorithm. A modified initialization scheme that leverages grid partitioning and oppositional-based learning is incorporated to produce an evenly distributed initial population across P-V curve. Additionally, a rank-based selection scheme is adopted to choose best half of population for subsequent global and local search modes. A modified global search method with fewer parameters is devised to rapidly identify approximated location of GMPP. Finally, a modified local search method using Perturb and Observe with adaptive step size method (P&O-ASM) is proposed to refine the near-optimal duty cycle and track GMPP with negligible oscillations. MPSO-HALS is implemented into low-cost microcontroller for real-time application. Extensive studies prove the proposed algorithm outperforms bat algorithm (BA), improved grey wolf optimizer (IGWO), conventional PSO and P&O, with convergence time shorter than 0.3s and tracking accuracy above 99% under different complex PSCs.