Abstract: In order to be seasoned with the regulation of reactor power during load change of marine nuclear power plant, a method based on running data learning was studied to compute demand reactor power, and experimentation was processed separately by support vector machine (SVM) and BP neural network. The results show that the method based on running data learning is more accurate than physical model method during the process of load rapid change, especially method based on SVM, which could get a smart predicting model only by one round training in a short time, and further more it can insure that the getting result is most optimal of all. Besides, this method could cope with situation that some input signal default, and therefore it would improve the stability, reliability and fault tolerance of computing.