An agent composed of data model and thermodynamic model for multi-component degradation identification of gas turbine online

To achieve accurate and efficient identification of gas turbine degradation states, a dual-drive agent combining a data-driven model and a thermodynamic model is proposed. A robust ensemble learning framework is first constructed by analyzing the coupling characteristics among gas turbine components, thereby enabling dimensionality reduction of the degradation identification space. The degradation state is then identified using the thermodynamic model in conjunction with the Rank Whale Decision Optimization Algorithm (RWDOA) within the reduced space. The combined use of data-driven learning and thermodynamic modeling significantly reduces the number of required training samples while enhancing identification accuracy. The proposed method achieves a classification accuracy exceeding 96.55 % for individual components, and the maximum identification error is limited to 0.0122. Compared with conventional model-based and data-driven approaches, the proposed dual-drive agent exhibits superior performance in both accuracy and stability, making it well-suited for online health monitoring of complex gas turbine systems.