Neural networks meet physics to predict supercapacitor lifespan: A breaking through method for energy storage monitoring

In our increasingly electrified world, supercapacitors have emerged as critical components in transportation and renewable energy systems, prized for their remarkable power density, cycling stability, and rapid charge-discharge capabilities. However, accurately predicting when these devices will degrade—and ultimately fail—has remained a significant challenge for engineers and energy systems designers. Traditional approaches have either relied too heavily on oversimplified physical models or required massive amounts of data for machine learning algorithms, creating a technological gap in the reliable monitoring of these essential energy storage devices.

Researchers from Beijing Institute of Technology introduced a hybrid approach that combines the best of both worlds: the interpretability of physics-based models with the adaptive power of neural networks. By developing a Physics-Informed Neural Network (PINN) built on Long Short-Term Memory (LSTM) architecture, they created a predictive system that achieves remarkable accuracy while requiring significantly less training data than conventional methods.

The main contribution of this research including:

(1) This model achieves a root mean square error (RMSE) of just 3 mF when predicting capacity degradation trajectories (against a rated capacity of 1 F)

(2) Remaining useful life predictions show an RMSE of only 269 cycles (compared to an average supercapacitor lifespan of 5,180 cycles)

(3) Compared to standard data-driven LSTM methods, this approach reduces prediction errors by 85% for degradation trajectories and 86.5% for remaining useful life

(4) When compared to traditional equation-based methods, the improvements are even more dramatic: 87.5% reduction in degradation prediction errors and 94.6% for lifespan prediction

Perhaps most impressively, this model maintains high accuracy even when trained on limited datasets—a crucial advantage in real-world scenarios where comprehensive historical data may not be available.

The implications of this research extend far beyond academic interest. The PINN approach opens new possibilities for:

(1) Electric Vehicle Battery Management: Implementation in onboard systems could provide drivers and fleet managers with precise estimates of energy storage health, optimizing charging strategies and preventing unexpected failures.

(2) Grid-Scale Energy Storage: Power utilities could deploy this technology to better manage large supercapacitor arrays in renewable energy installations, maximizing operational efficiency and planning maintenance schedules with unprecedented accuracy.

(3) Next-Generation Predictive Maintenance: The framework we've developed could be extended to other energy storage technologies, including various battery chemistries, creating a universal approach to energy storage health monitoring.

Future research directions include incorporating bidirectional LSTM or Gated Recurrent Unit (GRU) networks to enhance temporal feature extraction, as well as integrating more sophisticated electrochemical-thermal degradation models to improve performance under extreme operating conditions.

This research represents a paradigm shift in energy storage prognostics by successfully bridging the gap between purely data-driven approaches and physics-based modeling. By dynamically balancing physical constraints with neural network adaptability through Bayesian optimization, we've created a system that delivers high-precision predictions while maintaining physical consistency—all while requiring significantly less training data than conventional methods.

As the world transitions toward electrified transportation and renewable energy, the ability to accurately predict the health and remaining lifespan of energy storage components becomes increasingly crucial. Our physics-informed neural network approach provides a robust, adaptable solution that could help accelerate this global transition by making energy storage systems more reliable, predictable, and cost-effective.

Reference

Author: Lixin E ^a, Jun Wang ^a, Ruixin Yang ^a, Chenxu Wang ^a, Hailong Li ^b, Rui Xiong ^a

Title of original paper: A physics-informed neural network-based method for predicting degradation trajectories and remaining useful life of supercapacitors

Article link: https://www.sciencedirect.com/science/article/pii/S2773153725000416

Journal: Green Energy and Intelligent Transportation

DOI: 10.1016/j.geits.2025.100291

Affiliations:

^a Department of Vehicle Engineering, School of Mechanical Engineering, Beijing Institute of Technology, Beijing 100081, China

^b School of Business, Society and Engineering, Mälardalen University, Västerås 72123, Sweden