Engineers have long dreamed of delivering electricity through the air as reliably as through a wall socket.
Now, a research team from Chiba University has developed a fully numerical, machine learning-based design method that brings us a step closer to that future. By optimizing wireless power transfer (WPT) systems through real-world circuit modeling and genetic algorithms, the new approach achieves a stable output voltage and high efficiency—even when the power load fluctuates. The work could transform how we power everything from electric vehicles to implantable medical devices.
Why Load-Independence Matters in Wireless Power
Wireless power transfer works by sending energy through electromagnetic fields between two coils: one connected to a power source, the other to a device. But one of the biggest hurdles has been achieving “load independence”—a system that keeps delivering steady voltage no matter what device is on the receiving end.
Traditionally, engineers have relied on complex, hand-derived equations based on idealized assumptions to design WPT circuits. These models often fail when faced with real-world complexities like parasitic capacitance and unpredictable load changes. That’s where machine learning comes in.
Machine Learning Meets Power Electronics
The team, led by Professor Hiroo Sekiya of Chiba University’s Graduate School of Informatics, developed a fully numerical design method that ditches traditional equations in favor of differential equations and optimization algorithms. The circuit’s behavior is modeled step by step using real-world parameters, and a genetic algorithm iteratively fine-tunes the design to meet multiple goals:
- Maintain constant output voltage
- Maximize power-delivery efficiency
- Minimize total harmonic distortion
“We established a novel design procedure for a LI-WPT system that achieves a constant output voltage without control against load variations,” Prof. Sekiya said. “This is the first success of a fully numerical design based on machine learning in the field of power electronics research.”
Putting the Design to the Test
The researchers applied their technique to a class-EF WPT system—an advanced architecture that typically struggles when loads vary. In traditional designs, zero-voltage switching (ZVS), a key feature for energy efficiency, fails when the load strays from its ideal value. But the new system maintained ZVS and output voltage stability across a range of conditions.
In conventional designs, output voltage could vary by up to 18 percent depending on the load. The ML-designed system held this variation to under 5 percent. It also showed improved performance at lighter loads, thanks to accurate modeling of diode parasitic capacitance. At its rated point of 6.78 MHz, the system delivered over 23 watts of power with an efficiency of 86.7 percent.
Wider Implications and a Wireless Future
The team’s findings extend beyond this specific circuit type. Their work shows that machine learning can uncover optimal circuit configurations that traditional methods miss, including novel ways to reduce power loss in resonant filters.
“We are confident that the results of this research are a significant step toward a fully wireless society,” said Prof. Sekiya. “Due to load independence, the WPT system can be constructed in a simple manner, thereby reducing the cost and size.”
In other words, wireless power could soon become not just possible but practical—embedded in everything from home electronics to industrial systems—thanks to a marriage of circuit theory and artificial intelligence.
Journal and Citation
Journal: IEEE Transactions on Circuits and Systems I: Regular Papers
DOI: 10.1109/TCSI.2025.3579127
Article Title: ML-Based Fully-Numerical Design Method for Load-Independent Class-EF WPT Systems
Publication Date: June 18, 2025
ScienceBlog.com has no paywalls, no sponsored content, and no agenda beyond getting the science right. Every story here is written to inform, not to impress an advertiser or push a point of view.
Good science journalism takes time — reading the papers, checking the claims, finding researchers who can put findings in context. We do that work because we think it matters.
If you find this site useful, consider supporting it with a donation. Even a few dollars a month helps keep the coverage independent and free for everyone.