EPFL method identifies Li-ion battery parameters with under 1% error
New time-domain technique achieves under 1% error using standard BMS sensors.
Accurate estimation of a battery's internal parameters—especially low-frequency (LF) parameters that govern diffusion—is critical for state-of-health monitoring and grid services like primary frequency control. However, built-in battery management systems (BMS) sample voltage and current at low rates with limited precision, making it difficult to identify these parameters in the time domain. The paper introduces a modeling and identification framework that tackles this challenge head-on.
The core innovation lies in approximating the fractional constant phase element (CPE), which represents LF diffusion, as a high-order RC network with recursively defined values. The parameter estimation problem is then formulated as a constrained, non-convex least-squares problem in discrete state-space. The authors rigorously derive initialization strategies, bounds, and a procedure to select the optimal number of RC branches, improving robustness against noise. In a power system simulation where the battery provides primary frequency control, the method achieved average parameter errors below 1% under noise levels typical of commercial BMS sensors.
- Approximates fractional CPE using a high-order RC network with recursive definitions.
- Formulates estimation as a constrained non-convex least-squares problem in discrete state-space.
- Achieves sub-1% average estimation error even with noisy BMS measurements.
Why It Matters
Enables accurate battery state-of-health tracking and grid services without costly high-fidelity sensors, lowering barriers for storage deployment.