Fuzzy Logics as an Integral Part of Evolutionary Algorithms

Evolutionary algorithms play an important role for solving various optimization problems related to fuel management in reactor physics like core loading pattern optimization (LPO) or refueling. In general, all algorithms make a sample of solution candidates and evaluate the fitness of all candidates, and then the candidates with better fitness value are used to generate the next sample of solution candidates. In optimization algorithms, internal parameters [like population size, weighting factor in estimation of distribution algorithm (EDA) and population size, cross-over rate, etc., in Genetic Algorithm (GA)] have a stiffness problem as the value of these parameters is fixed at the first generation and is not being changed subsequently. However, the flexibility of changing the value of even one internal parameter during the generations will make the algorithm more efficient. In this paper we propose that fuzzy logics can be used in an innovative way to eliminate the stiffness problem related to internal parameters in evolutionary algorithms. As a test case, EDA for initial core LPO of the advanced heavy water reactor is chosen, and the use of fuzzy logics has shown a significant improvement in the algorithm’s performance. The appropriate value of weighting factor α in EDA has been predicted using fuzzy logics in each generation, and this has resulted in efficiency improvement of the algorithm. The improved methodology is expected to give better performance with other optimization algorithms, such as the GA or Ant Colony Optimization Algorithm, etc.