A research team at the Indian Institute of Technology (ISM) Dhanbad, led by Sunil K. Pradhan and Basab Chakraborty, has developed a grey wolf optimisation (GWO)-based hybrid regression model that significantly improves state-of-health (SOH) estimation for bipolar lead-acid batteries.
Random forest model optimised using a grey wolf algorithm
The work, published in Engineering Chemical Engineering, combines Lasso regression and support vector regression with a random forest model optimised using a grey wolf algorithm. The approach also integrates multiple electrochemical features extracted from partial charging data, including localised voltage area, sample entropy and fuzzy entropy.
The study was validated using six 6V bipolar lead-acid battery prototypes, fabricated using fused filament fabrication and multilayer lead substrates, and tested under controlled cycling conditions at 25°C.
Performance results show that the hybrid model achieved:
- Mean absolute error (MAE): below 1.02%
- Root mean squared error (RMSE): below 1.5%
- Relative error: under 6.2%, with 88% of results below 3.5%
Accuracy improved further with increased training data, with RMSE decreasing from 2.08% to 1.27% as dataset size increased.
The model outperformed several benchmark techniques, including Gaussian process regression, deep neural networks, recurrent neural networks and long short-term memory (LSTM) models.
A key advantage is the ability to estimate SOH using partial charging profiles, rather than full charge–discharge cycles. This makes the approach more applicable to real-world battery management systems, where full cycling data is rarely available.
The findings are particularly relevant for bipolar lead-acid batteries, which are gaining attention for their compact design, improved current pathways and higher power density compared with conventional configurations.
Photo: a grey wolf cub in a forest Credit: Shutterstock
