Energy-Efficient Transformer Framework for MPPT and Harmonic Mitigation in Multilevel Inverters

Abstract

Integration of multilevel inverters with renewable energy resourcesrequires high quality voltage synthesis and efficient power extraction under diverse operation conditions. Various traditional maximum power point tracking (MPPT) as well as inverter control models faces significant issues that results in poor convergence, oscillation susceptibility, as well as limited dynamic irradiances. All these limitations lead to degraded waveform quality, reduced energy capture, and increased stress on power electronic components. To mitigate these shortcomings, this paper proposes a novel Latent Attention Transformer (LAT) framework with MPPT and multilevel inverter control. The LAT predicts the optimal converter duty cycle, inverter switching angle corrections, and harmonic compensation factors in a unified manner. The differentiable multi-objective loss is formulated to maximize the PV power, minimize the total harmonic distortion, and penalize abrupt switching transitions. Training is performed on a larger dataset generated via MATLAB/Simulink cosimulation, incorporating different environmental and load conditions. The model is used on an edge co-processor with 8-bit quantization that ensures inference within the control cycle and effective integration with a DSPIC-based hardware system. A comprehensive analysis highlight the superiority of the proposed model by achieving maximum power point tracking (MPPT) efficiency of 97.2%, convergence times below 30 ms, and reducing inverter THD to 1.6% in simulation, compared to values exceeding 3% in other existing methods.