A Machine Learning Degradation Model for Electrochemical Capacitors Operated at High Temperature

Electrochemical capacitors (ECs) have only recently been considered as an alternative power source for telemetry sensors of drilling equipment for geothermal or oil and gas exploration. The lifecycle analysis and modelling of ECs is underrepresented in literature in comparison to other storage devices e.g. Li-ion batteries. This paper investigates the degradation of ECs when cycled outside the manufacturer-specified operating temperature envelope and proposes a machine learning-based approach for modelling the degradation. Experimental results show that end of life, defined as a 30% decrease in capacitance, occurs at 1,000 cycles when the environmental temperature exceeds the maximum operating temperature by 30%. The life-cycle test data is then used as an input to a Gaussian process regression (GPR) algorithm to predict the capacitance fade trend. The GPR is validated on a total of nine commercial cells from two different manufacturers, achieving an average root mean squared percent error of less than 2% and a mean calibration score of 93% when referenced to a 95% confidence interval. The model can be utilized to determine the EC degradation rate at a range of operating temperature values.