State-of-Charge Estimation of Lithium-ion Capacitors Combining Impedance Data and Deep Neural Network

Lithium-ion capacitors (LICs) are novel energy storage devices that are gaining popularity for combining the high energy density of lithium-ion batteries with the high power density of supercapacitors. Due to their hybrid nature, a signif-icant challenge lies in the estimation of their State of Charge (SOC). SOC estimation is crucial in practical applications of LICs, as accurate estimation improves longevity, performance, and dependability by enabling efficient energy management. This work presents two proposed schemes utilizing impedance data and deep neural networks (DNN) for SOC estimation of LICs. Electrochemical Impedance Spectroscopy (EIS) data is collected for a100F LIC across the entire SOC range, in 5% increments, at 25°C. This raw EIS data is directly fed as input to the DNN-1 for SOC estimation scheme-1, Further, an Equivalent Circuit Model (ECM) is identified using the raw EIS data. Subsequently, the parameters of this ECM are used as inputs for the DNN-2 in SOC estimation scheme-2. The performance of both the schemes are validated and compared. Although, both the schemes perform satisfactorily, having an overall Root Mean Square Error (RMSE) (taking the average RMSE of all the SOC points) of less than 1.3%, the DNN trained with raw impedance data moderately outperforms the DNN trained with ECM parameters, achieving an overall RMSE of less than 1%.