Prediction of physical fields for supercritical carbon dioxide turbine using deep learning method

Abstract

Supercritical carbon dioxide (S-CO₂) energy system has gained extensive attention recently, while design of its turbine is one of the most crucial tasks. In this research, we establish a data-driven model for the physical field prediction of a S-CO₂ turbine. This method can be applied to real-time prediction of physical fields in design and operation process of turbine. Firstly, a brief outline is presented, including previous computation efforts of S-CO₂ turbine and academic progress on applying deep learning in physical field prediction. Then, a specific S-CO₂ system is defined with details of blade profile geometry and operating conditions. Generation of the training data with CFD method is also covered. Next, the structure of the proposed neural network and its training strategies are formulated. To balance the prediction accuracy and the time cost, we build our model by basic multi-layer perceptron (MLP) model, with various depths of the hidden layers. Finally, accuracy of the predictive models under different training parameters are evaluated and compared to each other. The result demonstrates that the proposed framework is capable of predicting whole physical fields of the S-CO₂ turbine efficiently with overall mean square error (MSE) on test dataset as low as 3.187×10^-4, which implies its great potential in design and maintenance of S-CO₂ turbines.