A Core Loss Estimation Method Based on Data-driven Technology with Multi-head Attention Mechanism

Magnetic components play an important role in power electronic equipment, but there is still lack of accurate magnetic core loss estimation model. The data-driven methods are well-developed in recent years, and they are also more robust and accurate than conventional model-driven methods like Steinmetz Equation (SE) and improved Generalized Steinmetz Equation (iGSE). However, existing data-driven methods often focus on single excitation waveform like sinusoidal excitation waveform, which is not robust enough when facing complex excitation waveform with many high-frequency harmonic waveform. This paper presents a data-driven model based on attention mechanism which can capture the inner relationship within the input excitation waveform to present a more accurate core loss estimation results. And the comparison results with Multi-layer Perceptron (MLP) and Long Short-Term Memory(LSTM) models indicates that the proposed model has much more accuracy when facing complex excitation waveform. The proposed model can help to design magnetic components.