A Predictive Method for Impedance Estimation of Triboelectric Nanogenerators Based on a Gated Recurrent Unit Model

Abstract

Majorities of existing studies aim at investigating the influence of relevant parameters of triboelectric nanogenerators (TENGs) on their electrical output. However, few studies reported the relationship between these parameters and internal impedances of TENGs. In this work, we have firstly achieved accurate prediction for internal impedances of contact-separation TENGs (CS-TENGs) with different combinations of structural and motion parameters through a gated recurrent unit (GRU) model. Specifically, optimal impedances of CS-TENGs with different parameter combinations are obtained through experiments to construct a dataset. Furthermore, we build a novel GRU model with optimized hyperparameters for accurate prediction of internal impedances. Meanwhile, two other methods including back propagation (BP) neural network and convolution neural network (CNN) are used for comparison via two performance indexes of MAE and RMSE. Experimental results show that the built GRU model presents the lowest prediction error with MAE and RMSE being 0.45 and 0.52, respectively. Finally, we compare the extent of influence of these four parameters on internal impedances of CS-TENGs, showing an order as follows: contact area>speed>thickness>type. Summarily, this work is aimed at providing a convenient and reliable approach to achieve reasonable structural designs for TENGs according to the required external load in actuality.

Keywords: Triboelectric Nanogenerator, contact-separation mode, optimal matching impedance, Deep learning