A research team from Chiba University and its collaborators has developed a fully numerical, machine learning-based design method that significantly enhances the stability and efficiency of wireless power transfer (WPT) systems
This new approach overcomes some longstanding challenges in maintaining a constant output voltage across varying loads. This is an essential step towards practical and cost-effective wireless power solutions.
The challenge of load variability in WPT systems
Wireless power transfer systems transmit electricity without physical connectors, relying on electromagnetic fields between transmitter and receiver coils. While already applied in everyday devices like smartphones, electric toothbrushes, and sensors, WPT systems often suffer from voltage instability when the electrical load changes. This fluctuation can cause efficiency losses and degrade system performance, particularly in more demanding or variable use cases.
Typically, achieving load-independent (LI) operation requires precise tuning of circuit components such as capacitors and inductors. However, these values are normally calculated using idealised analytical equations that fail to account for real-world circuit behaviours, including parasitic elements and non-linear characteristics. As a result, the practical implementation of LI operation has remained a significant engineering challenge.
A fully numerical, AI-driven approach
To address these limitations, researchers led by Professor Hiroo Sekiya at the Graduate School of Informatics, Chiba University, developed a new machine learning-based method that uses numerical optimisation instead of analytical formulas. By modelling the WPT circuit with differential equations that capture real-world behaviour over time, the system’s performance can be evaluated and optimised step-by-step.
The team employed a genetic algorithm to adjust circuit parameters based on an evaluation function. This function considers output voltage stability, power-delivery efficiency, and harmonic distortion. Iteratively refining the design in this way allows the system to reach optimal steady-state operation without relying on idealised assumptions.
This fully numerical method shows a shift in WPT design, highlighting how machine learning and artificial intelligence can quicken the development of power electronics and lead to more robust and adaptable systems.
Experimental success
To validate their method, the researchers applied it to a class-EF WPT system, a design that combines a class-EF inverter with a class-D rectifier.
Conventional class-EF systems without load-independent optimisation can only maintain zero-voltage switching (ZVS) at specific load conditions. When the load deviates, ZVS is lost, reducing overall efficiency and stability.
The numerically optimised LI class-EF system maintained both ZVS and a steady output voltage across a wide range of load conditions. The researchers reported output voltage fluctuations of less than 5%, a significant improvement compared to the 18% variation observed in conventional systems.
The system achieved a high power-delivery efficiency of 86.7% at 6.78 MHz, delivering over 23 W of power. The design also showed consistent performance at light loads, with effective management of parasitic diode capacitance and stable power dissipation in the transmission coil.
A more practical wireless future
The implications of this research go beyond just performance improvements. The machine learning-driven design process simplifies circuit development, reduces cost and component sensitivity, and moves WPT technology closer to widespread commercial adoption.
The team hopes for applications in consumer electronics, medical devices, and industrial systems, where cable-free power delivery can provide convenience, safety, and flexibility.
As the world moves toward more innovative, more connected systems, advances like this show a future where wireless power is practical and also ubiquitous. With continued research and refinement, fully wireless infrastructure could become a standard feature of daily life within the next decade.