Deep Learning-Enhanced Parameter Extraction for Equivalent Circuit Modeling in Electrochemical Impedance Spectroscopy

Reliable and automatic parameter extraction in equivalent circuit modeling of electrochemical impedance spectroscopy (EIS) could be a challenge as the common circuit fitting method, complex nonlinear least-squares (CNLS), heavily depends on the initial guesses. To prevent the adjustment of the initial guess that demands extra time and experience, we propose employing a deep learning-based convolutional neural network (CNN) to perform the pre-fitting of the measured impedance spectrum. This approach not only facilitates the convergence dynamics of CNLS but also manifests a notable enhancement in parameter extraction fidelity, especially when benchmarked against conventional methodologies. The improvement of 25% in fitting success rate is demonstrated on an open-source impedance dataset by comparing to CNLS with random initials and the traditional stochastic methods including differential evolution and simulated annealing. Thus, we believe the proposed prefitting method can provide a useful tool for reliable parameter extraction with the uncertainty minimized to explore the underlying mechanism from EIS and automate this process for the analysis of a large amount of data.