Effective extraction and analysis of proton exchange membrane fuel cell (PEMFC) ageing characteristics is a prerequisite for long-term prediction of PEMFC ageing. To address the problem that it is difficult to comprehensively and effectively extract the reversible and irreversible ageing features data of PEMFC under dynamic operating conditions. In this paper, the equivalent inductor module is introduced on the basis of the classical equivalent circuit model to realize the accurate simulation of PEMFC dynamic behaviors. By further summarizing the activation polarization and ohmic polarization ageing laws, a PEMFC ageing behavior dynamic model is established, and the effective extraction of PEMFC reversible and irreversible ageing voltage component data under dynamic operating conditions is realized. Single-input recurrent convolutional neural networks cause the problem of prediction error accumulation due to the single dimension of training data. By building a multi-input recurrent convolutional neural network algorithm, the error accumulation of the prediction algorithm is reduced and the long-term ageing prediction accuracy of PEMFC is improved.