Optimization and discretization of dragonfly algorithm for solving continuous and discrete optimization problems

Bibi Amirah Shafaa, Emambocus (2024) Optimization and discretization of dragonfly algorithm for solving continuous and discrete optimization problems. Masters thesis, Sunway University.

Preview

Text OPTIMIZATION AND DISCRETIZATION OF DRAGONFLY.pdf - Accepted Version Download (14MB) | Preview

Abstract

Optimization is prevalent in almost all areas since a plethora of problems can be formulated as optimization problems. Hence, optimization algorithms, consisting of exact and heuristic methods, are crucial for a myriad of real-world applications. The exact methods tend to be computationally expensive and time-consuming. Hence, heuristic and metaheuristic algorithms are being increasingly favoured as they provide near-optimal solutions in a feasible amount of time. Swarm intelligence algorithms are metaheuristic algorithms inspired by the simple interactions of biological organisms in a population. Owing to their exploitation and exploration capabilities, swarm intelligence algorithms have a good performance in solving complex problems. The Dragonfly Algorithm (DA) is one which is inspired by the swarming behaviours of dragonflies during hunting and migrating in nature. It has been found to have a higher performance than multiple other swarm intelligence algorithms in various applications. However, despite having a good performance, it has certain limitations like a low exploitation phase. Furthermore, the original DA is only suitable for solving continuous optimization problems. Although there is a binary version of the algorithm, it cannot be directly used for solving discrete optimization problems like the Traveling Salesman Problem (TSP). Hence, in this project, the exploitation of the original DA is improved by using the stochastic hill climbing algorithm as a local search technique. The optimized continuous DA is applied for the training of Artificial Neural Networks (ANNs) so as to determine the optimal connection weights and biases for the network. The trained network is then applied to benchmark classification problems. Based on the experimental results, the optimized DA algorithm is a much better training algorithm for ANNs as compared to the usual gradient-descent backpropagation algorithm since the resultant ANNs trained by the optimized DA achieve higher accuracy. Moreover, the ANNs trained by the proposed optimized DA have a higher accuracy than those trained by the original DA when the performance of the algorithms are compared both in terms of iteration number and computational cost. The ANNs trained by the optimized DA also achieve higher accuracy than those trained by some other swarm intelligence algorithms. The optimized continuous DA is also applied for channel estimation in an optical communication system to determine the optimal channel coefficients. Based on the experimental results, the optimized DA is found to provide significantly better results than the original DA. Moreover, DA is adapted to be suitable for discrete optimization problems by adapting the equations of DA and by using the method of swap operators to update the position of the search agents. The adapted discrete DA is applied to a small TSP problem and it is able to provide the optimal solution for the small TSP problem. However, it is found to have low effectiveness for large TSP problems. Hence, the adapted discrete DA is optimized by improving the low exploitation phase. The steepest-ascent hill climbing algorithm is used as a local search technique to improve the exploitation of the adapted discrete DA. The optimized discrete DA is then applied to a TSP problem modelling a package delivery system in the area of Kuala Lumpur and to benchmark TSP problems. Based on the experimental results, the proposed optimized discrete DA has a higher effectiveness than the adapted discrete DA when the performance of the algorithms is compared in terms of both iteration number and computation cost. It also has a higher effectiveness than some other swarm intelligence algorithms. Furthermore, it provides optimal solutions for several benchmark TSP problems.

Actions (login required)

View Item