Researchers from the Singapore-MIT Alliance for Research and Technology (SMART), led by Sara Sameer and Wei Zhang, have developed a new method for estimating battery health that could significantly improve the management of batteries in electric vehicles and grid storage systems. The team’s work, published in the journal Nature Communications, introduces a model called Pace, which combines machine learning with battery physics to predict battery health more accurately and efficiently than existing methods.
Batteries are a crucial component in modern energy systems, but their performance degrades over time, which can lead to safety issues and increased costs. Accurate battery health estimation is essential for managing these systems effectively. The researchers developed Pace to address this challenge by integrating raw sensor measurements with battery physics features derived from an equivalent circuit model. This approach allows Pace to capture both the short- and long-term degradation patterns of batteries, leading to more accurate predictions.
Pace consists of three battery-specific modules. The first is a dilated temporal block that efficiently encodes temporal information from the battery’s usage. The second is a chunked attention block that models the context of the battery’s usage, allowing Pace to better understand the conditions under which the battery is operating. The third is a dual-head output block that fuses the short- and long-term degradation patterns of the battery, providing a more comprehensive estimate of its health.
The researchers tested Pace on a large public dataset and found that it outperformed existing models, achieving an average performance improvement of 6.5 and 2.0 times compared to the two best-performing baseline models. They also demonstrated Pace’s practical viability by deploying it in real-time on a Raspberry Pi, a low-cost, credit-card-sized computer.
The practical applications of Pace for the energy sector are significant. Accurate battery health estimation can help extend the lifespan of batteries, reduce costs, and improve the safety of energy storage systems. For electric vehicles, this could mean longer-range trips and lower maintenance costs. For grid storage systems, it could mean more reliable and efficient energy storage, which is essential for integrating renewable energy sources into the grid.
In summary, the researchers have developed a new method for estimating battery health that combines machine learning with battery physics. Pace has been shown to outperform existing models and has been demonstrated to be practical for real-time deployment. This work represents a significant advance in battery health analytics and has important implications for the energy sector.
Source: Nature Communications
This article is based on research available at arXiv.