Accurate Gas–Steam Combined Cycle Efficiency Prediction Based on Neural Network Model

(1) Background: To enhance the efficiency and minimize the energy consumption of combined cycle power plants (CCPPs), it is crucial to research gas–steam combined cycle (GSCC) performance prediction under various conditions. However, current studies focus more on the subsystems of GSCC, including simpler systems like gas turbines and steam turbines, lacking an overall perspective on the GSCC system as a whole. (2) Methods: this paper focuses on GSCC efficiency prediction, while employing continuous and fluctuating operational data from a CCPP. Specifically, variables from two symmetric gas turbines of the GSCC were employed as model inputs. Deep Neural Network, Simple Recurrent Neural Network, Long Short-Term Memory, and Gated Recurrent Unit (GRU) were tested. Furthermore, the GRU network was employed to evaluate the Plate Heat Exchanger (PHE) installation modification of the CCPP. (3) Results: GRU outperformed the other models, achieving a Mean Absolute Percentage Error of 0.855%. Utilizing multiple variables as model inputs provided the models better accuracy. The evaluation of the CCPP modification indicates that the PHE brought a maximum increase of 7.82 percentage points in combined cycle efficiency. (4) Conclusions: Recurrent Neural Networks, represented by GRU, are capable of predicting GSCC efficiency. Meanwhile, utilizing multiple inputs is essential to GSCC overall performance prediction. The research also proved the PHE to be effective in boosting GSCC thermal efficiency.