الفهرس 
CHAPTER (1) INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 116 |  |  |
| | | from | 116 | | |
Abstract
Nuclear fuel management optimization is one of the widely significant development activities in the field of nuclear reactor technology, that takes into consideration the reactor’s initial and posterior core configurations for generating the entire power within appropriate safety margins. It is related with the process of finding the optimum arrangement of fuel assemblies inside the reactor core. The optimum fuel assembly sites in the core are defined for the fresh and burned nuclear fuel assemblies available for core loading. The main goal of the fuel management is to maximize the lifetime of the reactor and using the minimum amount of fissile materials for the given fuel cycle length. So that, extra experiments can be performed with a specified fuel element inventory.
Finding the optimum configuration of fuel assemblies needs an enormous amount of calculations in traditional methods. Traditional methods are not qualified for accomplishing this amount of calculations in a rational time. For complicated problems like those associated with the nuclear fuel management optimization process, where a huge number of possible fuel arrangements exist, the use of Artificial Intelligence (AI) is a growing need. Various techniques have been evolved so as to automate the reloading process for the fuel assemblies in the reactor core. In the last years, Artificial Neural Networks (ANNs) were used efficiently as a progressive and committed tool for simulating different reactor physics parameters in nuclear applications.
In this thesis, a fast estimate tool is presented to solve the optimization fuel management problem. The tool automates the process of distributing the fuel assemblies inside the reactor core, and gets optimal final distribution. An algorithm based on Multilayer Perceptron Neural Networks (MPNNs), Apriori association rules and Particle Swarm Optimization (PSO) models was proposed for solving fuel loading pattern problem
The proposed algorithm provides a comprehensive analytic method for establishing an artificial neural network with self-regulation architecture; it finds an optimal number of hidden layers and their neurons, a less number of effective features of data set, and the most appropriate topology for internal connections. The developed 2-Dimensional neutronic diffusion code MUDICO-2D is used to get the required data needed for the training of the neural networks. The neural network development is recorded for a small modification in the core arrangement of the 10 MW IAEA benchmark LEU core research reactor.
The objective function was improved depending on two core safety parameters: the effective multiplication factor (Keff) and power peaking factor (Pmax). Optimal core configuration gets arrangements in which Keff is maximized, while Pmax value should be less than predefined value. Simulation results have been demonstrated the strength and the notability of the proposed algorithm comparing with recently existing ANN learning and classification algorithms. The proposed algorithm is much simpler to implement, achieves the highest classification accuracy with lowest number of nodes in hidden layers in less time possible, besides; it is able to extract the most effective features.