EUVIMEDEUVIMED entdecken
Crossref

A PHYSICS-INFORMED NEURAL NETWORK FRAMEWORK WITH POST-HOC SHAP INTERPRETABILITY FOR TRUSTWORTHY BATTERY STATE-OF HEALTH PROGNOSTICS

Surendra Reddy Pallamparthi, Shobhan Prasad Pujari

Phoenix: International Multidisciplinary Research Journal ( Peer reviewed High Impact Journal ) · 15.08.2024 · 2 (3) · 31-57

DOI: 10.65009/3neb8c42 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Phoenix: International Multidisciplinary Research Journal ( Peer reviewed High Impact Journal ), online erschienen am 15.08.2024. Verfasst von Surendra Reddy Pallamparthi und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The significance of Battery SOH (State-of-Health) prognostics for the enhancement of safety, reliability and operational efficiency of Lithium-Ion (Li-ion) battery systems, such as Energy Storage (ES) or Electric Vehicles (EV). Although conventional methods based on data might perform worse at inference, they can often aggravate the prediction task and cannot store the realistic battery degradation. In this paper, a neural network (NN) class known as PhysicsInformed Neural Network (PINN) model, along with a post-hoc interpretability approach using SHAP l methods, is proposed to confidently and interpretably predict Battery State-of-Health (SOH). This framework is based on a non-linear aging model built in a degradation-aware learning framework while maintaining the physics consistency of aging events and operability measurement of the battery. The battery is pre-processed using the battery voltage, current, temperature, internal resistance, battery state of charge and battery number of cycles and battery capacity data elements from the NASA Ames PCoE (Prognostics Centre of Excellence) Lithium-Ion Battery dataset. SHAP analysis is then used to numerically assess the importance of single operational factors and enhance the predictability transparency. Experimental analysis shows that the proposed estimation method yields 0.92 MAE, 1.34 RMSE, 1.31% MAPE, and an R² of 0.994, in addition to obtaining a Physics Consistency Score of 0.986 and a SOH Monotonicity Index of 0.989. The framework also provides battery health assessments with consistent feature level explanations, enabling reliable management of batteries with intelligent Battery Management Systems.

Zitieren

APA

Pallamparthi, S. R., & Pujari, S. P. (2024). A PHYSICS-INFORMED NEURAL NETWORK FRAMEWORK WITH POST-HOC SHAP INTERPRETABILITY FOR TRUSTWORTHY BATTERY STATE-OF HEALTH PROGNOSTICS. Phoenix: International Multidisciplinary Research Journal ( Peer reviewed High Impact Journal ), 2(3), 31-57. https://doi.org/10.65009/3neb8c42

Vancouver

Pallamparthi SR, Pujari SP. A PHYSICS-INFORMED NEURAL NETWORK FRAMEWORK WITH POST-HOC SHAP INTERPRETABILITY FOR TRUSTWORTHY BATTERY STATE-OF HEALTH PROGNOSTICS. Phoenix: International Multidisciplinary Research Journal ( Peer reviewed High Impact Journal ). 2024;2(3):31-57. doi:10.65009/3neb8c42

BibTeX
@article{Pallamparthi2024,
  author = {Pallamparthi, Surendra Reddy and Pujari, Shobhan Prasad},
  title = {A PHYSICS-INFORMED NEURAL NETWORK FRAMEWORK WITH POST-HOC SHAP INTERPRETABILITY FOR TRUSTWORTHY BATTERY STATE-OF HEALTH PROGNOSTICS},
  journal = {Phoenix: International Multidisciplinary Research Journal ( Peer reviewed High Impact Journal )},
  year = {2024},
  volume = {2},
  number = {3},
  pages = {31--57},
  doi = {10.65009/3neb8c42}
}
Kostenlos starten →