Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a sequential learning approach for parameter estimation in power converters using physics-informed neural networks (PINNs), addressing the challenge of identifying multiple interdependent parameters from time-domain data. By releasing gradients in a specific order (L, C, rL, rC) determined by sensitivity analysis, the staged PINN significantly improves parameter recovery accuracy and training efficiency compared to simultaneous training. The method achieves high accuracy even under component perturbations, demonstrating its robustness and potential for real-world applications.
Staged PINNs recover power converter parameters with 5% accuracy for L/C and 10% for ESR, outperforming simultaneous training by an order of magnitude in waveform RMSE.
A staged physics-informed neural network (PINN) is proposed for identifying the inductance, capacitance and parasitic resistances of a DC buck converter from time-domain data. Gradients are released in the order of L to C to rL to rC, a sequence derived from sensitivity analysis. The five-stage schedule recovers L and C to within 5% and the series ESR to 10%, while a three-stage variant attains the same accuracy 11% faster. An 84 factorial sweep confirms that the trained PINN remains accurate for simultaneous 8% perturbations of all four components, achieving a waveform RMSE of 2.6 x 10-2 over 5 ms-an order-of-magnitude better than a simultaneous-training baseline.