A deep learning method for fast prediction of turbine vane temperature fields under active modulation of gas turbine cooling air flow

Air-cooled turbines rely on cooling air extracted from the compressor to ensure safe operation. Under partial load conditions, hot component temperatures fall below design values, causing excess cooling and efficiency losses. Actively modulating turbine cooling air flow has been recognized as a key strategy to enhance efficiency, yet current open-loop systems lack precision due to limited understanding of vane temperature fields after modulation. To address this, this study proposes a deep learning method for fast prediction of turbine vane temperature fields under varying operating conditions and coolant volumes. A numerical model of a vane with complex cooling structures and a thermodynamic model of a three-shaft marine gas turbine are established. Using Latin hypercube sampling and co-simulation, a dataset of three-dimensional temperature fields is generated. A dual-branch conditional generative adversarial network with an attention mechanism is developed and trained using a combination of L1 and adversarial losses. The model achieves high accuracy, with a mean relative error of 0.1873 % and structural similarity of 0.9992. This approach enables fast, accurate predictions and reveals the effects of operating and coolant modulation conditions on vane temperatures.