NEURO-FUZZY BASED APPROACH TO ESTIMATE TURBINE-GENERATOR OUTPUT IN NUCLEAR POWER PLANTS

This paper presents a hybrid soft-computing modeling technique used to develop a turbine cycle model for the Maanshan Nuclear Power Plant (NPP) in Taiwan. The technique utilizes a neuro-fuzzy based approach to estimate turbine-generator output. First, operating data above the 95% load level from the plant's past three fuel cycles were collected and validated to serve as a baseline performance data set. Signal errors in new operating data were then detected and compared with the allowable range determined from the baseline data set. Finally, the variables most strongly influence turbine-generator output were selected as inputs for the neuro-fuzzy based turbine cycle model. After training and validation of key parameters, including main steam pressure, condenser backpressure, feedwater flow rate, and final feedwater temperature, the proposed model was used to estimate turbine-generator output. The effectiveness of the proposed neuro-fuzzy based turbine cycle model was demonstrated using plant operating data obtained from Manansan NPP. In addition, to assess the performance of the neuro-fuzzy based turbine cycle model, this study adopted a widely used commercial software program, PEPSE, for developing the thermodynamic turbine cycle of Maanshan NPP. Results show that the neuro-fuzzy based turbine cycle model is more reliable than the PEPSE turbine cycle model with the good estimation and the trend. Furthermore, the results of this study provide an alternative approach to evaluate thermal performance in nuclear power plants.