Application of computational intelligence methods to in-core fuel management

In this study, a computer program package has been developed which supports the in-core fuel management activities for pressurized water reactors, generates and recommends an optimum loading pattern to ensure safe and efficient reactor operation. A search for an optimum fuel loading pattern must be conducted in the space of several core parameters such as power distribution which is an excessively time consuming computational process. Global core calculation codes take a relatively long time to do the task. The time interval necessary for the iterative process was reduced by using an artificial neural network estimator for the calculations. In this way, it was possible to analyze more loading patterns in the same time interval and the probability of finding a desired optimum was increased. As a case study, the core of Almaraz Nuclear Power Plant of Spain, a pressurized water reactor, was modeled for the core calculation code system. The 2-group cross sections for the fuel assembly types were calculated and stored for later usage with the diffusion code. 2000 loading patterns were generated by placing fuel assemblies to random positions in the core, and for each pattern the power distribution and effective multiplication factor (keff) were calculated with the diffusion code. At the next stage, 500 of the loading patterns were introduced to the neural network as input data for the training process. The remaining 1500 patterns were used to validate the neural network implementation. It was shown that the neural network estimates the power distribution and the K effective within acceptable error limits. To complete system, a loading pattern generator was developed. This module consist of a set of rules and an algorithm that places the fuel assemblies to core positions. The neurol network estimated the power distribution and (keff) for the loading patterns that were generated by this module. The patterns that have a maximum power fraction lower than, and a minimum (keff) higher than reference values were stored as candidate optimum patterns. At the last stage of the work, an alternative loading pattern generator based on genetic algorithm method was developed. In this method, an initial loading pattern is improved by applying the genetic operators to obtain the optimum. The loading patterns obtained from the rule-based and the genetic algorithm methods were compared, and the genetic algorithm was shown to be more successful than the former. It was seen that, it is possible to automate in-core fuel management activities by applying artificial intelligence techniques