Development of supercritical CO2 turbomachinery off-design model using 1D mean-line method and Deep Neural Network

Abstract

Recently, supercritical CO2 (S-CO2) power system have received much attention due to their high efficiency and small size. S-CO2 exhibits a dramatic nonlinear property change near the critical point. Owing to this sensitivity, a bottleneck of S-CO2 system off-design analysis can be the turbomachinery off-design prediction. Conventionally, to reflect off-design performances of turbomachines within the system analysis, performance map with the correction method has been exploited. It is because computationally expensive to solve the Euler turbine equation for every iteration. However, due to the aforementioned behavior near the critical point, it is questionable if the method developed under air condition can still be valid for the S-CO2 system. The authors are proposing a method using Deep Neural Network (DNN) to build an S-CO2 turbomachinery off-design model. A statistical analysis revealed that the method showed 101 to 104 times better Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE) indices than those of the existing correction methods. Analysis from the system point of view was also carried out. The results of the pre-trained DNN S-CO2 turbomachinery off-design model and the correction methods were compared by analyzing the off-design steady-state performance of an S-CO2 simple recuperated cycle. The results showed that system off-design performance predictions can be significantly distorted with the conventional correction methodologies, and that it can be avoided through the developed DNN based S-CO2 turbomachinery off-design model.