Transformer prediction decomposition dual-driver for maximum power point tracking under partial shading conditions of photovoltaic power generation systems

Solar power, as a form of environmentally friendly and sustainable energy, has attracted considerable attention in the field of renewable energy power generation. However, the power curves of photovoltaic arrays are highly complex due to the inherent nonlinear characteristics of photovoltaic modules. To address the challenges of global maximum power point tracking under complex conditions, this study proposes a Transformer prediction decomposition dual-driver algorithm. The Transformer prediction decomposition dual-driver algorithm first utilizes a Transformer network to rapidly and accurately predict global maximum power point values of photovoltaic systems. Subsequently, the proposed algorithm integrates complete ensemble empirical mode decomposition with adaptive noise mode decomposition, fractional order proportional-integral-derivative control, and deep Q-network technology to achieve dual-mode control, thereby enhancing the stability of the control process. This study compares the Transformer prediction decomposition dual-driver algorithm with conventional algorithms and intelligent optimization algorithms under various shading conditions through simulation and hardware-in-the-loop experiments. Compared to other algorithms, the Transformer prediction decomposition dual-driver algorithm achieves a minimum increase in mean tracking efficacy of 0.1034% while concurrently reducing the mean tracking duration by no less than 0.0371 s. In physical experimental results, the proposed method exhibits minimal power disturbance compared to the reference algorithms.