Sequential Learning-Based Physics-Informed Neural Network Modeling for Parameter Estimation on a Power Converter

Samuela Rokocakau; Panagiotis Kakosimos; Giulia Tresca; Pericle Zanchetta

cs.LGOct 14, 2026

Sequential Learning-Based Physics-Informed Neural Network Modeling for Parameter Estimation on a Power Converter

Authors: Samuela Rokocakau, Panagiotis Kakosimos, Giulia Tresca, Pericle Zanchetta

Organizations: ABB Corporate Research Center, Väster˚as, Sweden · Dept. of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, Italy

Abstract

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.

Figures & tables

Fig. 1 : Buck converter with series inductor ESR

(r_{L})

, switch on-resistance

(r_{DS,\text{on}})

,and capacitor leakage

(r_{C})

explicitly modeled

Fig. 2 : PINN modeling workflow.

Scenario	Stages of learning
Parasitic inductor resistance ( $r_{L}$ only)	2
Parasitic capacitor resistance ( $r_{C}$ only)	2
Both $r_{L}$ and $r_{C}$	3
	5 *

TABLE I : Sequential-training stages required for each identification scenario

Component Parameter	True Values
Inductor, $L$	2 $m$ $H$
Capacitor, $C$	5 $\mu$ $F$
Parasitic Inductor Resistance, $r_{L}$	0.5 $\Omega$
Parasitic Capacitor Resistance, $r_{C}$	30 $k$ $\Omega$

TABLE II : Buck Converter Parameter True Values

Setting	Value / Rationale
Optimizer	Adam $\bigl[\beta_{1}=0.9,\;\beta_{2}=0.999\bigr]$
Learning rate	$1\times 10^{-3}$ during data pre-fit; $1\times 10^{-4}$ for every parameter-release stage
Batch size	Full record (5 001 time-samples, $5\,$ ms, $\Delta t=1\,\upmu$ s)
Gradient clip	$\lVert\nabla\theta\rVert_{2}\leq 1.0$ to avoid spikes when a parameter is unfrozen
Input scaling	$z$ -score on $i_{L}$ , $v_{O}$ , and duty ratio $S$ (mean 0, std 1)
Target scaling	Same $z$ -score applied to $\dot{i}_{L}$ and $\dot{v}_{O}$ ; data-loss weights $w_{i}=w_{v}=1$

TABLE III : Common hyper-parameters used in both training schedules

Fig. 3 : Stage 1 - True Data vs NN Simulation Plots

Fig. 4 : Stage 1 - True Data vs NN Simulation Plots

Offset (%)	$L$ ( $\mathrm{mH}$ )	$C$ ( $\mu\mathrm{F}$ )	$r_{L}$ ( $\Omega$ )	$r_{C}$ ( $\mathrm{k}\Omega$ )
–8	1.84	4.60	0.460	27.6
–6	1.88	4.70	0.470	28.2
–4	1.92	4.80	0.480	28.8
–2	1.96	4.90	0.490	29.4
+2	2.04	5.10	0.510	30.6
+4	2.08	5.20	0.520	31.2

TABLE IV : Parameter-sweep grid used for model-robustness analysis

Fig. 7 : RMSE Plots for Parameter Sweep Grid

Table 11

Fig. 8 : PINN (red) versus truth (blue) for the three best parameter sets.

Parameter Estimation

Parameter estimation studies how to infer unknown quantities in statistical, physical, and machine-learning models from observed data.

See all 144 papers

Neuromorphic Parameter Estimation for Power Converter Health Monitoring Using Spiking Neural Networks
Apr 17, 2026Hyeongmeen Baik, Hamed Poursiami, Maryam Parsa +1Parameter Estimation Predictive Maintenance
University of Wisconsin–Madison Madison, Wisconsin, USA · George Mason University Fairfax, Virginia, USA
Residual Physics-Informed Neural Networks for High-Fidelity BLDC Motor Modeling
Jul 10, 2026Haitham El-HussienyNeural Surrogate Modeling Physics-Informed ML
Department of Mechatronics and Robotics Engineering, Egypt-Japan University of Science and Technology (E-JUST), New Burg El-Arab City, Alexandria, Egypt
Physics-informed GNN for medium-high voltage AC power flow with edge-aware attention and line search correction operator
Sep 26, 2025Changhun Kim, Timon Conrad, Redwanul Karim +6Efficient Neural Network Inference Graph Attention Networks
Pattern Recognition Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany · Institute of Electrical Energy Systems, Friedrich-Alexander-Universität Erlangen-Nürnberg, Germany

Apr 17, 2026J/K move · Enter open · S save

Neuromorphic Parameter Estimation for Power Converter Health Monitoring Using Spiking Neural Networks

Hyeongmeen Baik, Hamed Poursiami, Maryam Parsa, Jinia Roy

University of Wisconsin–Madison Madison, Wisconsin, USA · George Mason University Fairfax, Virginia, USA

Parameter Estimation Predictive Maintenance Fault Detection Spiking Neural Networks Neuromorphic Computing Physics-Informed ML

Open paper AskPDF arXiv

Always-on converter health monitoring demands sub-mW edge inference, a regime inaccessible to GPU-based physics-informed neural networks. This work separates spiking temporal processing from physics enforcement: a three-layer leaky integrate-and-fire SNN estimates passive component parameters while a differentiable ODE solver provides physics-consistent training by decoupling the ODE physics loss from the unrolled spiking loop. On an EMI-corrupted synchronous buck converter benchmark, the SNN reduces lumped resistance error from

25.8\%

10.2\%

versus a feedforward baseline, within the

\pm 10\%

manufacturing tolerance of passive components, at a projected

{\sim}270\times

energy reduction on neuromorphic hardware. Persistent membrane states further enable degradation tracking and event-driven fault detection via a

+5.5

percentage-point spike-rate jump at abrupt faults. With

93\%

spike sparsity, the architecture is suited for always-on deployment on Intel Loihi 2 or BrainChip Akida.

Abstract

Parameter Estimation

Explore similar work

Neuromorphic Parameter Estimation for Power Converter Health Monitoring Using Spiking Neural Networks

Residual Physics-Informed Neural Networks for High-Fidelity BLDC Motor Modeling

Physics-informed GNN for medium-high voltage AC power flow with edge-aware attention and line search correction operator