LFP Battery SOC Estimation

Stanford Energy Controls Lab — 2025–Present Data Analysis, MATLAB, Machine Learning
EIS Nyquist plot analysis for LFP battery SOC estimation
Example Nyquist plot showing how a battery cell's impedance (its resistance to alternating current) varies with state of charge. These frequency-dependent shifts are the features we extract to estimate SOC using machine learning.

Lithium iron phosphate (LFP) batteries are increasingly deployed in EVs and grid storage, but their flat voltage plateau makes accurate state-of-charge (SOC) estimation notoriously difficult — real-world errors can reach up to 30%. This project uses electrochemical impedance spectroscopy (EIS) combined with machine learning to develop faster, more reliable SOC estimation methods suitable for real-time battery management systems.

My role focuses on extracting and analyzing impedance features from EIS Nyquist plots across multiple LFP cells to identify which measurement frequencies carry the most SOC information. This involves Spearman correlation analysis to find optimal single-frequency features, Gaussian Process Regression with Shapley value analysis to rank frequency importance collectively, and cross-cell generalization studies to determine whether impedance-based features transfer reliably between different battery cells.