Performance of Estimation of distribution algorithm for initial core loading optimization of AHWR-LEU

•EDA has been applied to optimize initial core of AHWR-LEU.•Suitable value of weighing factor ‘α’ and population size in EDA was estimated.•The effect of varying initial distribution function on optimized solution was studied.•For comparison, Genetic algorithm was also applied. Population based evolutionary algorithms now form an integral part of fuel management in nuclear reactors and are frequently being used for fuel loading pattern optimization (LPO) problems. In this paper we have applied Estimation of distribution algorithm (EDA) to optimize initial core loading pattern (LP) of AHWR-LEU. In EDA, new solutions are generated by sampling the probability distribution model estimated from the selected best candidate solutions. The weighing factor ‘α’ decides the fraction of current best solution for updating the probability distribution function after each generation. A wider use of EDA warrants a comprehensive study on parameters like population size, weighing factor ‘α’ and initial probability distribution function. In the present study, we have done an extensive analysis on these parameters (population size, weighing factor ‘α’ and initial probability distribution function) in EDA. It is observed that choosing a very small value of ‘α’ may limit the search of optimized solutions in the near vicinity of initial probability distribution function and better loading patterns which are away from initial distribution function may not be considered with due weightage. It is also observed that increasing the population size improves the optimized loading pattern, however the algorithm still fails if the initial distribution function is not close to the expected optimized solution. We have tried to find out the suitable values for ‘α’ and population size to be considered for AHWR-LEU initial core loading pattern optimization problem. For sake of comparison and completeness, we have also addressed the initial core optimization of AHWR-LEU by using Genetic algorithm (GA). In GA too, similar dependence on population size and initial distribution function is observed. However, by increasing the population size, the results in GA optimization improved drastically.