Degradation Prediction of the Hydrogen Fuel Cells Based on the Decoupled Echo State Network with Reservoir Predictive Mechanism

In the data-driven prediction methods, the echo state network (ESN) model could realize the prediction of proton exchange membrane fuel cells of degradation. Aiming at the problem of low prediction accuracy, a decoupled ESN (DESN) with the lateral inhibition based on reservoir predictive (DESN-RP) mechanism is proposed in this paper. By improving the structure of ESN and inhibiting the influence of other neurons and sub-reservoirs on the activated neurons, the preliminary decoupling of DESN is realized. The reservoir predictive (RP) mechanism accelerates the network learning of useful information and improves the prediction by strengthening the competition of activated neurons and inhibiting other neurons. It could further weaken the coupling of neurons and reduce the redundant information of the internal state. In general, DESN-RP could enhance feature representation, increase sparsity, reduce the fitting risk, and reinforce the generalization ability of the network. It was proved that DESN-RP improved the accuracy of long-term prediction of the degradation of PEMFC under steady-state, quasi-dynamic, and dynamic conditions.