Data-driven modeling approach of heavy-duty gas turbine with physical constraint by MTGNN and Transformer

Abstract

Aiming at the complex nonlinear and uncertain multi-variable coupling problems of heavy-duty gas turbine, a data-driven heavy-duty gas turbine modeling method combined with physical constraint is proposed. Firstly, by combining the advantages of MTGNN that focuses on local features and Transformer that extracts global information, an improved network is constructed to model heavy-duty gas turbine with full operating conditions and wide load variation. Secondly, the physical constraint related to the mechanism of heavy-duty gas turbine is added to eliminate the unreasonable output in the proposed heavy-duty gas turbine model, which improves the reliability and interpretability of the proposed modeling approach. Then, the Lyapunov function is utilized to prove the convergence of the proposed modeling approach. Finally, the actual datasets of heavy-duty gas turbine are used to carry out relevant emulation and comparison experiments. The results show that the proposed modeling approach is superior to other typical network models, which verifies the accuracy and practicability of the proposed modeling approach.