Physics-Informed Neural Networks for Joint Estimation of Multiparameters and Coil Misalignment in IPT Systems

In Inductive Power Transfer (IPT) systems, real-time perception of mutual inductance and load parameters is critical for regulating operating states and monitoring system conditions. This paper proposes a Physics-Informed Neural Network (PINN)-based multi-parameter joint identification method. By embedding physical constraints of impedance angle-mutual inductance-load relationships as domain knowledge, PINN is guided to learn system dynamics. Performance is further enhanced by integrating the Convolutional Block Attention Module (CBAM), which outperforms other attention mechanisms with very low convergence loss. Experiments on a 10 kW low-voltage, high-current IPT platform validate the high-precision identification: relative errors of 7.7% (coil displacement), 2.1% (mutual inductance), 2.0% (leakage inductance), and 1.8% (load). The method operates without system modification or additional hardware, making it suitable for online parameter identification and real-time monitoring in high misalignment and wide-load-range IPT applications.