Constant Current/Voltage Charging Control for Communication-Free IPT Systems via Network-Based Deep Transfer Learning

Parameter identification-based control strategies are considered the preferred solution for achieving constant current and constant voltage charging control in communication-free inductive power transfer systems. However, identification errors in mutual inductance, load resistance and other parameters can affect control accuracy. To improve control accuracy and response speed, a communication-free control strategy based on neural networks and deep transfer learning is proposed. This strategy eliminates the need to identify parameters like mutual inductance and load resistance. Only a few measured data are required to train the network model offline, a trained model can online estimate output voltage/current will be obtained. By combining with a controller, constant voltage/current charging control can be achieved under conditions of real-time variations in mutual inductance and load resistance. The experimental results show that the proposed control strategy achieves a static error of only 1.5% and a response time of no more than 24ms. Compared to the parameter identification-based control strategy, the proposed strategy demonstrates lower static error, shorter response time, and a wider dynamic range.