Degradation Prediction of PEMFCs Based on Discrete Wavelet Transform and Decoupled Echo State Network

Area of Science:

• Energy Systems Engineering
• Computational Science
• Materials Science

Background:

• Accurate prediction of proton exchange membrane fuel cell (PEMFC) degradation is vital for effective maintenance and health monitoring.
• Dynamic operational conditions and limitations of current forecasting methods hinder precise PEMFC degradation predictions.

Purpose of the Study:

• To develop an advanced predictive model for PEMFC degradation that enhances accuracy, particularly for long-term forecasting.
• To integrate discrete wavelet transform (DWT) with a novel decoupled echo state network (DESN-Z) for improved PEMFC performance prediction.

Main Methods:

• Utilized discrete wavelet transform (DWT) for noise reduction and feature extraction from PEMFC operational data.
• Introduced a novel decoupled echo state network with a decreasing inhibition mechanism (DESN-Z) to refine prediction accuracy.
• Combined DWT and DESN-Z (DDESN-Z) to leverage enhanced feature representation and network generalization.

Main Results:

• The DDESN-Z model effectively mitigates noise and overfitting, leading to improved feature representation and sparsity.
• Demonstrated significantly enhanced precision in long-term PEMFC degradation predictions across static, quasi-dynamic, and fully dynamic operational scenarios.
• The lateral inhibition mechanism in DESN-Z expedites information acquisition and refines predictions by managing neuron interconnectivity.

Conclusions:

• The DDESN-Z approach offers a robust solution for accurate long-term PEMFC degradation prediction.
• This method enhances the reliability of PEMFC health monitoring and maintenance planning.
• The study highlights the potential of combining signal processing techniques with advanced neural networks for complex system prognostics.