Application of Artificial Neural Networks in the Optimization of Feedwater Heater Models

Abstract The traditional lumped parameter model for feedwater heaters is insufficient to accurately predict changes in heat transfer performance caused by factors such as aging and fouling during power plant operation. This study established an operational database consisting of 68 million datasets and employed it to train an artificial neural network model. This neural network model establishes a mapping relationship between operating time, thermal hydraulic conditions, and heat transfer efficiency correction coefficients. Subsequently, the neural network model was employed to optimize the heat transfer calculation process of the lumped parameter model, and a new composite model was formed after integration. Finally, the optimization effect of the model was tested using operational data. The results show that compared with the lumped parameter model, the composite model exhibits higher and more stable computational accuracy at different operation times. This means that the composite model can, to some extent, reflect the changes in heat transfer performance of the feedwater heater during operation.