An identification method based on growing and pruning radial basis function network (GAP-RBFN) is presented for modeling an accelerator driven system (ADS). Compared with traditional neural networks, GAP-RBFN could automatically adjust the number of hidden neurons to find a suitable network structure by using growing and pruning strategies. In addition, an extended Kalman filter (EKF) algorithm is adopted to update network parameters of neurons in GAP-RBFN, which has a rapid convergence speed during the training process. A numerical calculation code named ARTAP (ADS Reactor Transient Analysis Program) is used to generate data for training GAP-RBFN. After GAP-RBFN is trained by the data, an identification model for ADS is established. The simulation results obtained from the GAP-RBFN model are compared with those obtained from a recurrent neural network (RNN) model. It is shown that the GAP-RBFN model not only has higher prediction accuracy than the RNN model, but also has faster computation speed than the numerical calculation code. Owing to its accuracy, simplicity and fast computation speed, the proposed GAP-RBFN method can be used to model the ADS reactor.

This content is only available via PDF.
You do not currently have access to this content.