Predicting Receiver Characteristics without Sensors in an LC-LC Tuned Wireless Power Transfer System Using Machine...

Area of Science:

• Electrical Engineering
• Power Electronics
• Machine Learning Applications

Background:

• Wireless Power Transfer (WPT) systems face challenges in efficiency, cost, and maintenance.
• Existing Receiver (Rx)-sensorless WPT synchronization methods lack consistent accuracy and availability.
• Machine Learning (ML) offers a promising approach to enhance WPT performance.

Purpose of the Study:

• To replace traditional Transmitter (Tx)-Rx communication with ML-based prediction.
• To utilize Tx-side parameters for predicting load and coupling coefficients in an LC-LC tuned WPT system.
• To improve the efficiency and accuracy of WPT systems through ML integration.

Main Methods:

• Developed two ML models to predict load and coupling coefficients.
• Utilized current and voltage features from the Tx-side for model training.
• Employed an extra trees regressor for predicting WPT system characteristics.

Main Results:

• The extra trees regressor achieved high accuracy in predicting load and coupling coefficients.
• Coefficients of determination were 0.967 for load and 0.996 for coupling.
• Mean absolute percentage errors were as low as 0.11% for load and 0.017% for coupling.

Conclusions:

• ML effectively replaces Tx-Rx communication in LC-LC tuned WPT systems.
• The proposed ML approach significantly improves prediction accuracy for WPT parameters.
• This method offers a cost-effective and efficient solution for WPT system optimization.