Data-Driven Electrolyzer Modeling: Adaptive Model Considering Operating Conditions using K-means Clustering

Abstract

This paper proposes a data-driven method for modeling electrolyzers at the cell level that takes into account operating conditions such as pressure, temperature, and current. To achieve this, operating conditions were categorized into optimal clusters using the K-means clustering algorithm. A deep neural network (DNN) was used to map the complex nonlinear input-output relationships arising from the electrolyzer's thermodynamic and electrochemical reactions. The study used a dataset of experimental data obtained from various specifications and operating conditions installed in different regions, with the goal of creating an adaptive electrolyzer model. The results showed that the proposed model outperformed physical-based and data-driven models that did not consider operating conditions in all evaluation indices. Specifically, the modeling error was MSE 0.15V/cell, RMSE 12.15mV/cell, MAE 8.14mV, and RE 0.49%. Therefore, the proposed model is suitable for energy grid research such as digital twins in future studies.