Tuesday 11 March 2025
The quest for more accurate battery life predictions has been a long-standing challenge in the field of electric vehicles and energy storage. A new study published in a scientific journal has made significant progress in this area by developing a novel approach that uses feature engineering and a specific type of neural network to predict the remaining useful life (RUL) of lithium-ion batteries.
The researchers, who hail from institutions in South Korea and the United States, started by collecting data from NASA’s Prognostics Center of Excellence on 20 features related to battery performance, including current, voltage, temperature, and time. They then used two different methods – partial correlation coefficient (PCC) and SHAP values – to identify the most important features for predicting RUL.
The results showed that PCC-based feature selection was effective in identifying key parameters for individual batteries, but the importance of these features varied significantly across different cells. In contrast, SHAP analysis consistently identified a set of features that were important across multiple batteries, despite differences in cell-to-cell variability.
The researchers then used these features to train a DLinear neural network, a type of recurrent neural network (RNN) designed specifically for time-series data with linear trends. The results were impressive: the DLinear model outperformed both long short-term memory (LSTM) and transformer models in predicting RUL, even when trained on fewer cycles.
One of the key advantages of the DLinear approach is its ability to capture the underlying trend in battery degradation over time. This is particularly important for lithium-ion batteries, which degrade rapidly during the early stages of their lifespan but more slowly as they age.
The study’s findings have significant implications for the development of condition-based maintenance (CBM) strategies for electric vehicles and other applications where battery health is critical. By predicting RUL with greater accuracy, manufacturers can reduce waste, minimize downtime, and optimize battery performance over its entire lifecycle.
The researchers also identified two key features – F2 and F11 – that were consistently important across multiple batteries. These features are related to variance in measured voltage during discharging and temperature measurements, respectively. Including these features in the DLinear model improved prediction accuracy even further.
While there is still much work to be done in developing more accurate battery life predictions, this study represents a significant step forward in the field. As the demand for electric vehicles continues to grow, the ability to predict RUL with greater precision will become increasingly important for ensuring the reliable and efficient operation of these systems.
Cite this article: “Predicting Remaining Useful Life in Lithium-Ion Batteries Using Novel Neural Network Approach”, The Science Archive, 2025.
Lithium-Ion Batteries, Electric Vehicles, Battery Life Prediction, Feature Engineering, Neural Networks, Dlinear Model, Rul, Condition-Based Maintenance, Cbm, Battery Health.
