The pellet-cladding interaction (PCI) is one of the major issues in fuel rod during the power change operation in light water reactors (LWR). The frequent power transient will lead to the PCI failure, which results in the radioactive leakage. Unfortunately, the failure cannot be detected on time during operation. For solving this problem, the PCI results is calculated and predicted before the power transient, and the results is used for optimizing the operation strategy. As everyone knows, the calculation of the fuel is a cumbersome work, which is not fit for online evaluation. At first, we present a PCI online prognosis model in this paper by the method of radial basis function neural network (RBFNN). Based on this online model, an optimal operation strategy is present then. The PCI evaluation results is treated as the feedback of the strategy, which will go to the power control system to decide to carry out the power plan or to modify it. The presented operation strategy used the fast online prognosis model to predict the PCI status, which is used as a feedback signal to the power control system. The optimal operation strategy is tested by the experimental data from the references, and the results demonstrate its effectiveness.