面向水下无线充电系统的机理-数据联合参数辨识混合方法

Accurate identification of key parameters is crucial for closed-loop control and system stability in underwater wireless power transfer (UWPT) systems. However, the absence of communication and severe harmonic distortion from eddy currents in sea environments present significant challenges to identification results. This article proposes a hybrid mechanism- and data-driven method for accurate joint identification of mutual inductance and load. First, to resolve the linear dependence and parameter coupling inherent in typical LCC–S compensated UWPT systems, a multifrequency excitation strategy is introduced to extract informative features for joint identification. Second, an artificial neural network is pretrained on mechanism-based data to construct the source domain model (SDM). Moreover, transfer learning fine-tunes the SDM with limited experimental data to obtain a target domain model (TDM), enhancing model generalization under varying conditions. Finally, an experimental platform with 1 kW/85 kHz is constructed to validate the proposed method. Experimental results and three different typical underwater conditions are captured and collected, which demonstrates that the proposed mechanism and data-driven joint parameter identification method can achieve quick and accurate identification results with prediction error below 3% and response time within 2.72 ms.