• Nuclear Engineering and Technology
• /
• v.54 no.10
• /
• pp.3864-3877
• /
• 2022

We propose a new approach called PESA (Prioritized replay Evolutionary and Swarm Algorithms) combining prioritized replay of reinforcement learning with hybrid evolutionary algorithms. PESA hybridizes different evolutionary and swarm algorithms such as particle swarm optimization, evolution strategies, simulated annealing, and differential evolution, with a modular approach to account for other algorithms. PESA hybridizes three algorithms by storing their solutions in a shared replay memory, then applying prioritized replay to redistribute data between the integral algorithms in frequent form based on their fitness and priority values, which significantly enhances sample diversity and algorithm exploration. Additionally, greedy replay is used implicitly to improve PESA exploitation close to the end of evolution. PESA features in balancing exploration and exploitation during search and the parallel computing result in an agnostic excellent performance over a wide range of experiments and problems presented in this work. PESA also shows very good scalability with number of processors in solving an expensive problem of optimizing nuclear fuel in nuclear power plants. PESA's competitive performance and modularity over all experiments allow it to join the family of evolutionary algorithms as a new hybrid algorithm; unleashing the power of parallel computing for expensive optimization.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.34 no.3
• /
• pp.125-136
• /
• 2009

The container packing problem Is one of the traditional optimization problems, which is very related to the knapsack problem and the bin packing problem. In this paper, we deal with the quadratic multiple container picking problem (QMCPP) and it Is known as a NP-hard problem. Thus, It seems to be natural to use a heuristic approach such as evolutionary algorithms for solving the QMCPP. Until now, only a few researchers have studied on this problem and some evolutionary algorithms have been proposed. This paper introduces a new efficient evolutionary algorithm for the QMCPP. The proposed algorithm is devised by improving the original network random key method, which is employed as an encoding method in evolutionary algorithms. And we also propose local search algorithms and incorporate them with the proposed evolutionary algorithm. Finally we compare the proposed algorithm with the previous algorithms and show the proposed algorithm finds the new best results in most of the benchmark instances.

• The KIPS Transactions:PartB
• /
• v.11B no.2
• /
• pp.213-220
• /
• 2004

Evolutionary algorithms are well-suited for multi-objective optimization problems involving several. often conflicting objective. Pareto-based evolutionary algorithms, in particular, have shown better performance than other multi-objective evolutionary algorithms in comparison. Recently, pareto-based evolutionary algorithms uses a density information in fitness assignment scheme for generating uniform distributed global pareto optimal front. However, the usage of density information is not Important elements in a whole evolution path but plays an auxiliary role in order to make uniform distribution. In this paper, we propose an evolutionary algorithms for multi-objective optimization which assigns the fitness using pareto dominance rank and density weighting, and thus pareto dominance rank and density have similar influence on the whole evolution path. Furthermore, the experimental results, which applied our method to the six multi-objective optimization problems, show that the proposed algorithms show more promising results.

• Journal of Korean Institute of Industrial Engineers
• /
• v.31 no.1
• /
• pp.10-19
• /
• 2005

The minimum spanning tree (MST) problem is one of the traditional optimization problems. Unlike the MST, the degree constrained minimum spanning tree (DCMST) of a graph cannot, in general, be found using a polynomial time algorithm. So, finding the DCMST of a graph is a well-known NP-hard problem of importance in communications network design, road network design and other network-related problems. So, it seems to be natural to use evolutionary algorithms for solving DCMST. Especially, when applying an evolutionary algorithm to spanning tree problems, a representation and search operators should be considered simultaneously. This paper introduces a new tree representation scheme and a genetic operator for solving combinatorial tree problem using evolutionary algorithms. We performed empirical comparisons with other tree representations on several test instances and could confirm that the proposed method is superior to other tree representations. Even it is superior to edge set representation which is known as the best algorithm.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.26 no.2
• /
• pp.47-58
• /
• 2001

The purpose of this study is to present methods for determining the k most vital arcs (k-MVAs) in shortest path problem (SPP) using evolutionary algorithms. The problem of finding the k-MVAs in SPP is to find a set of k arcs whose simultaneous removal from the network causes the greatest increase in the shortest distance between two specified nodes. Generally, the problem of determining the k-MVAs in SPP has been known as NP-hard. Therefore, to deal with problems of the real world, heuristic algorithms are needed. In this study we present three kinds of evolutionary algorithms for finding the k-MVAs in SPP, and then to evaluate the performance of proposed algorithms.

• Architectural research
• /
• v.4 no.1
• /
• pp.33-38
• /
• 2002

The objective of this study is the development of discrete optimum design algorithms which is based on the genetic algorithms. The developed algorithms was implemented in a computer program. For the optimum design, the objective function is the weight of space trusses structures and the constraints are stresses and displacements. This study solves the problem by introducing the genetic algorithms. The genetic algorithms consists of genetic process and evolutionary process. The genetic process selects the next design points based on the survivability of the current design points. The evolutionary process evaluates the survivability of the design points selected from the genetic process. The efficiency and validity of the developed discrete optimum design algorithms was verified by applying the algorithms to optimum design examples.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1998.10a
• /
• pp.420-426
• /
• 1998

Evolutionary algorithms are probabilistic optimization algorithms based on the model of natural evolution. Recently the efforts to improve the performance of evolutionary algorithms have been made extensively. In this paper, we introduce the research for improving the convergence rate and search faculty of evolution algorithms by using reinforcement learning. After providing an introduction to evolution algorithms and reinforcement learning, we present adaptive genetic algorithms, reinforcement genetic programming, and reinforcement evolution strategies which are combined with reinforcement learning. Adaptive genetic algorithms generate mutation probabilities of each locus by interacting with the environment according to reinforcement learning. Reinforcement genetic programming executes crossover and mutation operations based on reinforcement and inhibition mechanism of reinforcement learning. Reinforcement evolution strategies use the variances of fitness occurred by mutation to make the reinforcement signals which estimate and control the step length.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.1 no.1
• /
• pp.6-12
• /
• 2001

Evolutionary algorithms(EAs) have been successfully applied to many combinatorial optimization problems of various engineering fields. Recently, some comparative studies of EAs with other stochastic search algorithms have, however, shown that they are similar to, or even are not comparable to other heuristic search. In this paper, a new hybrid evolutionary algorithm utilizing a new local heuristic search, for combinatorial optimization problems, is presented. The new intelligent local heuristic search is described, and the behavior of the hybrid search algorithm is investigated on two well-known problems: traveling salesman problems (TSPs), and quadratic assignment problems(QAPs). The results indicate that the proposed hybrid is able to produce solutions of high quality compared with some of evolutionary and simulated annealing.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.37 no.3
• /
• pp.39-55
• /
• 2012

This paper presents an endosymbiotic evolutionary algorithm (EEA) to solve both problems of line balancing and model sequencing in a mixed-model two-sided assembly line (MMtAL) simultaneously. It is important to have a proper balancing and model sequencing for an efficient operation of MMtAL. EEA imitates the natural evolution process of endosymbionts, which is an extension of existing symbiotic evolutionary algorithms. It provides a proper balance between parallel search with the separated individuals representing partial solutions and integrated search with endosymbionts representing entire solutions. The strategy of localized coevolution and the concept of steady-state genetic algorithms are used to improve the search efficiency. The experimental results reveal that EEA is better than two compared symbiotic evolutionary algorithms as well as a traditional genetic algorithm in solution quality.

• Proceedings of the KSME Conference
• /
• 2000.11a
• /
• pp.699-704
• /
• 2000

The present paper deals with a multiobjective optimization method based on the co-evolutionary genetic strategy. The co-evolutionary strategy carries out the multiobjective optimization in such way that it optimizes individual objective function as compared with each generation's value while there are more than two genetic evolutions at the same time. In this study, the designs that are out of the given constraint map compared with other objective function value are excepted by the penalty. The proposed multiobjective genetic algorithms are distinguished from other optimization methods because it seeks for the optimized value through the simultaneous search without the help of the single-objective values which have to be obtained in advance of the multiobjective designs. The proposed strategy easily applied to well-developed genetic algorithms since it doesn't need any further formulation for the multiobjective optimization. The paper describes the co-evolutionary strategy and compares design results on the simple structural optimization problem.