Machine learning-based design enables more efficient wireless power transfer

Wireless Power Transfer (WPT) systems use electromagnetic fields to transmit electrical energy from a power source to a load without a physical connector or wire. The idea dates back to the 1890s, when Nikola Tesla made it famous for its famous wireless energy transfer.

Today, WPT systems are widely used for powering smartphones for the Internet of Things, electric toothbrushes and wireless sensors. A typical WPT system has a transmitter coil connected to a power source. This transmitter converts the power supplied into an electromagnetic field. The electromagnetic field is received by the receiver coil that supplies power to the electrical device.

Recently, load-independent (LI) operations have attracted attention to stabilizing the output voltage and maintaining zero voltage switching (ZVS) even when the load changes. However, achieving LI operations usually requires very accurate circuit component values, such as inductors and capacitors, calculated using complex analytical equations. These equations are often based on idealized assumptions and do not capture actual complexity.

To overcome these limitations and improve power delivery efficiency, a research team led by Professor Hirao Sekiya of the Faculty of Information Studies at Chiba University, Japan, proposed a machine learning-based design method for designing LI-WPT systems. This study was conducted in collaboration with Fukuda and Yutaro Yutaro University of komiyamo. Dr. Wenki Zhu from the Faculty of Electrical Engineering, Tokyo University of Science. Dr. Kihirojima from the Computer Information Science Department at Sojo University. This study was published online in the journal IEEE Transaction I on Circuits and Systems: A Regular Paper June 18, 2025.

In this approach, WPT circuits are written using differential equations that capture how voltages and currents evolve over time within a system, taking into account real-world component characteristics. These equations are resolved numerically in stages until the system reaches steady state.

The evaluation function evaluates the performance of the system based on key goals such as output voltage stability, power delivery efficiency, and total harmonic distortion. The genetic algorithm updates system parameters to improve evaluation scores, and is repeated until the process achieves a desired load-independent operation.

As Professor Sekiya explains, “We have established a new design procedure for LI-WPT systems that achieve constant output voltage without control over load fluctuations. We consider load independence to be an important technique for social implementation of WPT systems.

To test their method, the team applied it to a Class-EF WPT system that combines Class-F inverters and Class-D rectifiers. A traditional class-EF inverter without LI operation can only maintain ZV at its rated operating point. Changes in load will result in loss of ZV and reduced efficiency. In contrast, the LI version designed by the team stabilized ZV and output voltage even when loads changed.

In traditional systems with LI inverters, the output voltage can fluctuate by 18% when the load changes. In contrast, systems designed in a fully numerical way keep this variation below 5%, indicating significantly greater stability. By accurately explaining the effects of diode parasitic capacitances, the new approach has improved performance at optical loads.

Detailed power loss analysis revealed that transmission coils dissipate roughly the same amount of power under different load conditions, thanks to the system's ability to stabilize the output current. At a rated operating point, the LI Class-EF WPT system achieved 86.7% power delivery efficiency at 6.78 MHz and delivered output power of over 23 W.

Going forward, researchers believe that the meaning of this work goes beyond WPT. “We believe that the results of this study are an important step towards a fully wireless society. Furthermore, the operation of LI allows WPT systems to be built in a simple way, thereby reducing costs and scale.

More broadly, this design method shows that artificial intelligence and machine learning could change the way powered electronics are designed towards the future of automated circuit design.