Abstract
Under partial shading (PS) condition, the P-V curve becomes more complex where many peaks (one global maximum peak [GMP] and many other local maximum peaks [LMPs]) are generated. This GMP changes with time under a time-variant PS; this is called dynamic GMP. Conventional particle swarm optimization (PSO) can track the GMP under the same PS effectively. Nevertheless, it cannot track the dynamic GMP because all particles will be concentrated at the first GMP caught. In addition, using PSO as a maximum power point tracker (MPPT) technique suffers from obvious power oscillations in the steady state. In this paper, the PSO technique is improved to make it able to follow the dynamic GMP under time-invariant PS. In addition, a novel deep recurrent neural network (DRNN) is introduced to track the dynamic GMP under time-variant PS. A detailed comparison between DRNN and improved PSO is introduced, analyzed, and discussed. DRNN performs well compared with the improved PSO in terms of dynamic GMP tracking with almost zero steady-state oscillation, tracking speed, accuracy, and efficiency.