Accurate prediction of green hydrogen production based on solid oxide electrolysis cell via soft computing algorithms.

The solid oxide electrolysis cell (SOEC) presents significant potential for transforming renewable energy into green hydrogen. Traditional modeling approaches, however, are constrained by their applicability to specific SOEC systems. This study aims to develop robust, data-driven models that accurately capture the complex relationships between input and output parameters within the hydrogen production process. To achieve this, advanced machine learning techniques were utilized, including Random Forests (RFs), Convolutional Neural Networks (CNNs), Linear Regression, Artificial Neural Networks (ANNs), Elastic Net, Ridge and Lasso Regressions, Decision Trees (DTs), Support Vector Machines (SVMs), k-Nearest Neighbors (KNN), Gradient Boosting Machines (GBMs), Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machines (LightGBM), CatBoost, and Gaussian Process. These models were trained and validated using a dataset consisting of 351 data points, with performance evaluated through various metrics and visual methods. The dataset's suitability for model training was confirmed using the Monte Carlo outlier detection method. Results indicate that within the dataset and evaluation framework of this study, ANNs, CNNs, Gradient Boosting, and XGBoost models have demonstrated high accuracy and reliability, achieving the largest R-squared scores and the smallest error metrics. Sensitivity analysis reveals that all input parameters significantly influence hydrogen production magnitude. Game-theoretic SHAP values underline current and cathode electrode conditions as critical factors. This research determines the performance of machine learning models, particularly ANNs, CNNs, Gradient Boosting, and XGBoost, in predicting hydrogen production through the SOEC process. The outcomes of this paper can provide a certain reference for related research and applications in the hydrogen production field.