MLP-Based Intelligent Control Signal Estimation for Buck–Boost DC–DC Converters Using SHAP Analysis

This work introduces a complete machine-learning-driven scheme to make proper predictions and analysis of a nonlinear buck-boost DC-DC converter duty-cycle. The model of multilayer perceptron (MLP) is built based on the important electrical characteristics, such as input voltage, feedback voltage, inductor current, capacitor voltage, which allows making strong data-based predictions without using typical averaged or small-signal models. Through performance analysis, it is shown that the performance of the process is very much tracked accurately between the predicted and measured duty-cycle values with very few deviations of errors as confirmed by time domain plots, error histograms and correlation analysis. Moreover, SHAP-based explainability methods are used to measure the level of importance of features and understand the behavior inside the model, which show that the most important parameters affecting the duty cycle regulation is$V_{\text {in }}$and$i_{L}$. The findings verify that the suggested machine learning framework offers a valid, interpretable, and computationally efficient answer that can be used in the context of real-time power-electronic control, intelligent monitoring and renewable-energy scenarios.