| Title |
Physics-based and Data-driven Digital Twin Technology: Focusing on Battery Energy Storage Systems, Electrolyzers, Hydrogen Fueling Facilities, and Electric Vehicle Charging Facilities |
| Authors |
Munseok Chang ; Sungyu Hwang ; Sungwoo Bae |
| DOI |
http://doi.org/10.5207/JIEIE.2024.38.6.452 |
| Keywords |
Battery energy storage system; Data-driven modeling; Digital twin; Electric vehicle charging facility; Electrolyzer; Hydrogen fueling facility; Physics-based modeling |
| Abstract |
Digital twin technology creates models that reflect reality in a virtual environment, enabling analysis of both the physical and virtual worlds. It is a technology that allows for the understanding of past and present system operations and the prediction of future states. This study presents a method for modeling four types of facilities using both physics-based and data-driven modeling techniques through digital twins. First, for the battery energy storage system, a physics-based equivalent circuit modeling method was applied to simulate the chemical reactions and electrical relationships of the battery cells through internal parameters. This approach was then expanded to the battery pack level to evaluate the accuracy of the digital twin about the terminal voltage. For the electrolyzers, a physics-based equivalent circuit modeling method was initially used to reflect the electrochemical and thermodynamic characteristics of the electrolysis mechanism. This approach was then expanded from the cell level to the system level, and a data-driven deep neural network model was additionally developed to evaluate the accuracy of the digital twin regarding output voltage. The hydrogen fueling facility was modeled using a physics-based gas equation of state that considers the energy and enthalpy of hydrogen gas. Additionally, a data-driven random forest model was employed to determine the accuracy regarding the internal pressure of the hydrogen tank. Lastly, the electric vehicle charging facility was modeled using a physics-based equivalent circuit model to simulate the nonlinear charging profile, and a data-driven support vector machine model was built to estimate charging demand patterns. The performance of the digital twin modeling for the four facilities was evaluated based on the normalized mean absolute error metric, yielding results of 1.26%, 2.50%, 2.93%, and 5.47%, demonstrating highly favorable outcomes. |