Machine Learning Based Accurate Modeling of Rectenna Nonlinear Behavior

The nonlinear behavior of rectenna is a critical bottleneck restricting the efficiency of wireless power transfer. Most existing works utilize a nonlinear model derived from analyzing the I–V characteristics of diodes and taking a Taylor expansion. In this paper, we employ two advanced machine learning models, LASSO regularized Polynomial Regression (LASSO-PR) and Segmented Polynomial Regression (S-PR), to derive a more precise and adaptive nonlinear model. Besides, we design a single-diode rectenna in ADS to compare the accuracy of these models. The training results demonstrate that the existing nonlinear model is only suitable for small input power levels, whereas the S-PR based model can accurately characterizes the nonlinear behavior across the entire input power range, reducing the output dc voltage prediction error by 82%.