Abstract
Direct methanol fuel cells (DMFCs) offer a promising solution for clean electricity generation, particularly in small electronics and remote auxiliary power units. However, optimizing their efficiency and performance is challenging due to the complex interactions between various factors. Here, we present a novel approach that integrates experiments with machine learning to model and predict the performance of these fuel cells using atomically dispersed platinum group metal (PGM)-free catalysts at the cathode. Our machine learning models, trained on diverse input parameters, allow for the comprehensive optimization of DMFC performance prior to fabrication and testing. Through extensive experimental validation, we demonstrate that this data-driven approach accurately predicts key performance metrics, such as maximum power output and polarization curves. By combining our models with interpretable game-theory methods, we provide deep insights into the factors governing fuel cell performance, ultimately paving the way for the design of scalable and efficient DMFC technologies.
| Original language | English |
|---|---|
| Article number | 235758 |
| Journal | Journal of Power Sources |
| Volume | 626 |
| DOIs | |
| State | Published - Jan 15 2025 |
Funding
The database with sixty samples was used to train seven machine learning algorithms, including linear (Linear) and second-order polynomial (Poly) algorithms, support vector regression (SVR), k-nearest neighbors (KNN), gradient boosting trees (GBT), random forest (RF), and multilayer perceptron artificial neural networks (ANN). These algorithms were chosen for their ability to handle both linear and nonlinear relationships in the data [58]. While simpler models like linear regression offer easy interpretation, more complex models like random forest and neural networks can capture intricate patterns in the data. Other algorithms could also be explored, but these seven already provide us with sufficient diversity to quickly evaluate which approaches provide the best predictive accuracy for DMFC performance. All models were implemented using the SCIKIT-LEARN and details about each of these algorithms can be found in the computational package's documentation [64]. Initially, the models were trained considering the peak power density as the target performance, which enabled the later employment of explainable computational methods (e.g., permutation importance analysis, Shapley additive explanatory approaches) for a deeper understanding of how different variables affect the final performance of DMFCs. In the second stage, we also considered an ANN model to surrogate the relation between the four input variables and the full polarization curve. Unless noted otherwise, we randomly split the initial database in a 75 %\u201325 % training-to-testing size ratio. We optimized the hyperparameters of the SVR, KNN, GBT, RF, and ANN models by minimizing the root-mean-squared error (RMSE) cost function averaged over twenty random shuffle split cross-validation sets, each containing 20 % of the training data (i.e., the training, cross-validation, and testing sets correspond to 60 %, 15 %, and 25 % of the entire database). Table 1 shows the hyperparameters considered for optimization and their final converged values following a brute-force grid search over the hyperparameter space. We used the default value as implemented in SCIKIT-LEARN for the hyperparameters not listed in the table.The work was funded by the U.S. Department of Energy (DOE), Energy Efficiency and Renewable Energy, Hydrogen and Fuel Cell Technologies Office through Electrocatalysis Consortium (ElectroCat). The authors thank the ElectroCat DOE program managers, Dimitrios Papageorgopoulos, David Peterson, McKenzie Hubert, and William Gibbons. This work was authored in part by Los Alamos National Laboratory operated by Triad National Security, LLC under US DOE contract no. 89233218CNA000001. This manuscript has been authored by UT-Battelle, LLC, under contract DE-AC05-00OR22725 with the US Department of Energy (DOE). The US government retains and the publisher, by accepting the article for publication, acknowledges that the US government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript or allow others to do so, for US government purposes. DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan). The work was funded by the U.S. Department of Energy (DOE), Energy Efficiency and Renewable Energy, Hydrogen and Fuel Cell Technologies Office through Electrocatalysis Consortium (ElectroCat). The authors thank the ElectroCat DOE program managers, Dimitrios Papageorgopoulos, David Peterson, McKenzie Hubert, and William Gibbons. This work was authored in part by Los Alamos National Laboratory operated by Triad National Security, LLC under US DOE contract no. 89233218CNA000001. This manuscript has been authored by UT-Battelle, LLC, under contract DE-AC05-00OR22725 with the US Department of Energy (DOE). The US government retains and the publisher, by accepting the article for publication, acknowledges that the US government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript or allow others to do so, for US government purposes. DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan ( http://energy.gov/downloads/doe-public-access-plan ).
| Funders | Funder number |
|---|---|
| DOE Public Access Plan | |
| Hydrogen and Fuel Cell Technologies Office | |
| Los Alamos National Laboratory | |
| ElectroCat DOE | |
| Electrocatalysis Consortium | |
| U.S. Department of Energy | DE-AC05-00OR22725, 89233218CNA000001 |
| U.S. Department of Energy |
Keywords
- Data-driven optimization
- Direct methanol fuel cells
- Electrocatalysis
- Machine learning
- Oxygen reduction reaction
- PGM-Free catalysts
