Abstract. In this paper, a feed forward artificial neural network (ANN) is used to predict the effective multiplication factor (k ef f ), an indication of the reactivity of a nuclear reactor, given a fuel Loading Pattern (LP). In nuclear engineering, the k ef f is normally calculated by running computer models, e.g. Monte Carlo model and finite element model, which can be very computationally expensive. In case that a large number of reactor simulations is required, e.g. searching for the optimal LP that maximizes the k ef f in a solution space of 10 10 to 10 100 , the computational time may not be practical. A feed forward ANN is then trained to perform fast and accurate k ef f prediction, by using the known LPs and corresponding k ef f s. The experiments results show that the proposed ANN provides accurate, fast and robust k ef f predictions.Keywords: Feed forward neural network, Nuclear reactor, k ef f prediction.
Nuclear Reactor Reactivity EstimationNuclear Reactors are loaded with fission fuel elements to generate energy for electricity or neutrons for medical/physics research. A crucial part of nuclear reactor operations is to decided how to load the fuel, in another word, to design the core Loading Pattern (LP), to improve the performance of the reactor, indicated by an estimation of the core reactivity and/or some other key measures, e.g. power and temperature distributions.A reactor core normally contains a number of loadable positions (or fuel channels), from tens to hundred of thousands, to be filled with designated fuel elements. An LP can be regarded as an ordering assignment of n items to m positions. In this case, n is the number of fuel elements, and m is number of fuel channels. When the size of the reactor is large, the number of possible LPs need to be examined before loading the reactor can be very large.Given a designed LP, it is desired that some key parameters of the reactor can be estimated. Such calculations are done by building a computer model and