Data-Driven and Physics-Guided Neural Approaches for LCR Circuit Modeling: A Comparative Analysis
Year: 2026
Type: book
Abstract: This study compares a data-driven RNN–LSTM predictor with a physics-guidedneural approximation for the transient charge response of an LCR circuit. Thereference trajectory is obtained from a Sumudu Transform–Adomian Decom-position Method (Sumudu–ADM) treatment of a fractional integro-differentialcircuit model. The two neural models solve different learning tasks: the RNN–LSTM performs one-step sequence prediction from historical state variables,whereas the physics-guided network maps time to charge while penalizing adifferential-equation residual. Importantly, the implemented physics residual usesa first-order effective damping term rather than a discrete Caputo operator; itis therefore not claimed to be a genuine fractional-order PINN. The reportederror measures are interpreted as task-specific approximation results rather thanevidence of the inherent superiority of either architecture. The study furtheridentifies limitations associated with synthetic reference data, the validation ofthe truncated Sumudu–ADM series, and the non-equivalent information suppliedto the two learning frameworks. These observations motivate future work usingindependently validated fractional solvers, genuine nonlocal fractional residuals,leakage-free multi-step forecasting, and measured circuit data.
Authors: Nimai Sarkar, Tanneeru, Gopisairam
Institution: Vellore Institute of Technology University
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204160792