Hybrid Knowledge and Data-Driven Hydrogen Trading for Renewable-Dominated Hydrogen Refueling Stations

This paper proposes a hybrid knowledge and data-driven predict-then-optimize paradigm for green hydrogen (H<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub>) trading among renewable-dominated hydrogen refueling stations (HRSs). Firstly, a data-driven H<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> load forecasting method is formulated where the key influencing features are captured by XGBoost and the Informer algorithm with encoder and decoder processes is utilized to generate the predicted time series of hydrogen load. Then, a bi-level hybrid knowledge and data-driven H<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> trading model with rolling horizon optimization is proposed to determine the optimal trading quantity of H<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> and dynamically optimize the transportation routes for the traded H<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> based on the cell transmission model and traffic state. Moreover, a fully distributed solution algorithm is developed to decompose the complex multi-period H<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> trading problem into local electricity and hydrogen dispatch subproblems of HRSs for efficiently obtaining the optimal H<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> trading amount. Comparative studies have demonstrated the superior performance of the proposed methodology on the improvement of the distributed renewable energy accommodation and economic benefits for HRSs.