Determining battery SoC using Electrochemical Impedance Spectroscopy and the Extreme Learning Machine

Abstract

Much effort has been made in recent years to accurately determine battery state-of-charge (SoC) and state-of-health (SoH). Electrochemical impedance spectroscopy (EIS) is well-established for parameter identification; however EIS has traditionally been a laboratory procedure. With the recent prevalence of low-cost DSPs, it has become feasible to use EIS in online applications. This paper focuses on implementing EIS using a DC/DC converter topology commonly found in renewable energy applications. An AC ripple voltage is injected into the battery by modulating the PWM duty cycle, then the current and phase-shift response is analyzed to determine the frequency-dependent impedance. Voltage and current sensing devices have been developed so that the technique can be implemented on a TI F2833 DSP. EIS is performed at set intervals during entire charge cycles on test batteries in order to produce a data-driven model. Regression is performed using the Extreme Learning Machine (ELM) neural-network algorithm. The derived model is then verified by predicting the SoC of a battery used as a test sample.