New data-driven model for estimating the weather-dependent thermodynamic performance of nuclear power plants

Abstract

This research developed a new data-driven model to estimate the impact of condenser pressure, a function of the temperature of the tertiary coolant and thus the surrounding weather, on the thermodynamic performance of a nuclear power plant (NPP). Artificial Neural Network (ANN) was utilized to construct the data-driven model. Considering the wide variety of thermodynamic cycles and plants components involved in the NPPs, two simplified thermodynamic models were developed, one applicable for water-cooled nuclear reactor-based power plants and the other one for metal and gas-cooled reactor-based plants. These thermodynamic cycles were employed to generaterandom data for training two separate ANN models, each trained with 20,000 datasets. To make the simplified thermodynamic models and ANN models generalized i.e., applicable for all the nuclear power plant options available throughout the world, a self-calibrating feature was added to the models through global optimization algorithms where the isentropic efficiencies of the turbines would be calibrated to match the rated output power of the NPP. Two evolutionary algorithms, Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) were considered for the purpose. To observe which algorithm has superior calibration performance, a simplified thermodynamic model for VVER-1200 from the available literature was calibrated using both the algorithms. Results revealed that GA can ensure better calibration of the model compared to PSO. To evaluate the sensitivity of the calibration process to the selection of rated condenser pressure of a NPP, calibration of the model for VVER-1200 was done for two condenser pressures, 4kPa and 7kPa. Results indicated that the calibrated models for both cases have almost identical predictive accuracies. A comparative analysis between the simplified thermodynamic models and the ANN models was performed to realize the justification for constructing the data-driven models. It was observed that all these models had similar predictive performances but the self-calibration time for the ANN models are around 10-20 times less than the simplified thermodynamic models, indicating a significant reduction in computational costs. The developed models were utilized to observe the changes of efficiencies, output powers and condenser thermal loads of different Gen III, III+ and IV reactor based NPPs with the change of condenser pressure after calibrating them with data from Advanced Reactors Information System (ARIS). The condenser pressure was varied in the range 4-15kPa. It was observed that the metal and gas-cooled reactor-based NPPs are comparatively less affected by increased condenser pressure than the water-cooled ones.