Physics-Informed Neural Networks to predict the Power Transmission of Electric Road Systems

Résumé fourni par la source

A Physics-Informed Neural Network (PINN) is employed to predict the transmission efficiency of an Electric Road System (ERS) which is a promising technique to boost electric mobility. The ERS consists of transmitter coils integrated into the road and receiver coils underneath the electric vehicles (EVs). The technique enables the dynamic charging of EVs via inductive power transfer (IPT). An established method to improve the transmission efficiency of the IPT is the use of ferrite structures (e.g. plates behind the coils) to guide the magnetic field. In COMSOL Multiphysics, a multiscale 3D Finite Element Method (FEM) simulation of the ERS is implemented, where Maxwell’s equations are solved numerically to model the IPT process. The transmitter coils are excited by an alternating current which allows the IPT to be formulated in the frequency domain. The numerical model is used to generate training data for the deep neural network (DNN). For the proposed PINN, the dimensions of the ferrite plate and the relative positions of the transmitter and the receiver coils are varied to generate training data. After the training and validation phase, Maxwell’s equations are encoded into the PINN, which is then capable of predicting key ERS parameters (e.g. transmission efficiency) in real time. The accuracy of the PINN was evaluated on a randomly selected set of test data.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.