Abstract: The selection of the pressurized water reactor core refueling scheme is directly related to the safety and economy of nuclear power plant operation, which is a time-consuming and laborious work. In order to efficiently and accurately select feasible core loading schemes, a layered neural network model is proposed to predict the key parameters of the core loading scheme, namely the cycle length and maximum enthalpy rise factor. Our proposed method designs a double-layer hidden layer network structure. By selecting the appropriate weight initialization method, activation function, adaptive learning rate and optimizer, a large amount of engineering refueling data is learned to obtain the core key parameter prediction model. Our proposed method learns key parameter features separately and avoids mutual interference learning, thus improving the overall prediction accuracy of the model. Numerical experiments also show that our proposed method has higher prediction accuracy and stronger robustness than the classical DNN neural network model, and the layered neural network model also has accurate prediction ability on different types of data.