Identifying Equivalent Circuits of Electrochemical Impedance Spectroscopy Using CNN-LSTM

Electrochemical impedance spectroscopy (EIS) is widely utilized in electrochemical studies, and equivalent circuits are usually the first and crucial step in analyzing impedance data. In this study, an artificial intelligence (AI)-driven approach combining convolution neural network (CNN) and long short-term memory (LSTM) is proposed to identify equivalent circuits. The simulated data were preprocessed to construct a dataset for the CNN to extract the characteristic EIS feature. The stacked LSTM network was used to detect the in-depth sequence features of the EIS, and then the equivalent circuit recognition result was obtained. Initial experiments indicate that compared with standalone CNN, LSTM, and their improved models, the accuracy of CNN-LSTM model reaches 93.27%, and the accuracy, recall and F1 score improved by at least 1.11%,0.87% and 0.66%, respectively. Finally, the model was further validated using impedance data from lithium-ion batteries, demonstrating consistent probability distributions of equivalent circuits compared to circuit fitting results. These findings confirm the effectiveness of CNN-LSTM model in identifying equivalent circuit models for EIS data.