A Hybrid Prognostic Method for Proton-Exchange-Membrane Fuel Cell with Decomposition Forecasting Framework Based on...

Area of Science:

• Electrochemical Engineering
• Materials Science
• Artificial Intelligence

Background:

• Proton-exchange-membrane fuel cells (PEMFCs) face durability and reliability challenges hindering commercialization.
• Accurate prognostic methods are crucial for reducing maintenance costs and extending PEMFC operational lifetime.
• Existing methods often struggle with long-term degradation prediction and capturing complex aging behaviors.

Purpose of the Study:

• To propose a novel hybrid prognostic method for accurate long-term voltage degradation prediction in PEMFCs.
• To enhance the reliability and lifespan estimation of PEMFCs for commercial deployment.
• To improve the efficiency and accuracy of prognostic models using advanced machine learning techniques.

Main Methods:

• A decomposition forecasting framework is employed, separating voltage data into calendar and reversible aging components using Locally Weighted Scatterplot Smoothing (LOESS).
• An Adaptive Extended Kalman Filter (AEKF) predicts the reversible aging component, while a Long Short-Term Memory (LSTM) neural network predicts the calendar aging component.
• A genetic algorithm-based automated machine learning approach is used to optimize the LSTM model for improved prediction accuracy and efficiency.

Main Results:

• The hybrid method accurately predicts long-term voltage degradation in PEMFCs.
• The proposed approach demonstrates superior performance compared to single model-based or purely data-driven methods.
• Remaining useful life estimation is achieved by summing the predicted aging components.

Conclusions:

• The hybrid prognostic method offers a robust solution for predicting PEMFC voltage degradation.
• This approach can significantly contribute to the commercial viability of PEMFC technology by improving reliability and reducing operational costs.
• The integration of decomposition, AEKF, LSTM, and genetic algorithms provides a powerful tool for fuel cell prognostics.