Algoritmo genético aplicado à identificação de parâmetros de um conversor boost

Abstract

With the increased penetration of renewable sources in the electrical grid, particularly solar photovoltaic generation, the use of DC-DC converters has become a fundamental part of the efficient integration of these systems into the grid. However, in several practical applications, the physical parameters and accurate models of these converters are not completely known, hindering rigorous simulations and a proper understanding of the dynamic behavior of these systems. In this context, this work proposes a methodology for the parametric identification of Boost converters operating in continuous conduction mode, based on the integration of small-signal modeling, pseudo-random binary signal (PRBS) excitation, and genetic algorithms. First, the theoretical basis of the Boost converter is developed, analytically deducing the transfer function Gvg(s) using the state space averaging method. This deduction establishes the dependence between the physical parameters resistor (R), inductor (L), capacitor (C) and duty cycle (D), and the dynamic response of the converter. Subsequently, the experimental methodology is implemented in the MATLAB/Simulink environment, introducing small PRBS-type disturbances in the input voltage, which makes it possible to obtain the transfer function experimentally through differential spectral analysis. The genetic algorithm is configured with real encoding, with each chromosome responsible for simultaneously estimating the four parameters of the converter, using as an objective function the minimization of the normalized mean square error between the magnitudes of the theoretical and experimental transfer functions. The robustness of the proposed technique is evaluated considering three progressive scenarios of increasing complexity: ideal components with narrow search limits (±1%), inclusion of non-idealities (equivalent series resistances - ESR), and expansion of search limits to larger ranges (±5% and ±10%). The results obtained demonstrated good accuracy in parametric identification, with relative errors within the established ranges, validating the effectiveness and reliability of the proposed methodology. This study contributes to the field of parametric identification applied to power electronics, offering a robust, versatile, and replicable computational approach with the potential to be used in diagnostics, controller self-tuning, and predictive maintenance of DC-DC converters, and can also be extended to other topologies and industrial contexts.