Prediction of Key Core Parameter of PWR by Adaptive BP Neural Network

The adaptive BP (back propagation) neural network method was used, by realizing mathematical modeling of core loading mode, adaptive selection of network nodes, learning rate and random gradient descent search, and three key parameters including effective multiplication factor, component power peak factor and rod power peak factor of core fuel refueling for Qinshan Ⅱ PWR were quickly and accurately predicted. Compared with the traditional method, the BP neural network algorithm saves a lot of calculation consume. Numerical experiment results show that maximum relative error is less than 2% for conditions beyond the training data, so the algorithm has good robustness and high reliability, which makes an important exploration for the further application of artificial intelligence algorithm in nuclear industry.