Development of Random Walks Strategy-Based Dandelion Optimizer and Its Application to Engineering Design Problems

The objective of this paper is to enhance the swarm-based metaheuristic dandelion optimizer (DO) algorithm by incorporating various strategies to address early convergence and mitigate the risk of becoming trapped in local optima. This integration is intended to yield optimal and favorable outcomes for real-world optimization problems. To achieve this objective, a novel hybrid algorithm called the random walks dandelion optimizer (RW-DO) was introduced. This new algorithm addresses the limitations of the dandelion optimizer DO when handling optimization problems by incorporating at random walks strategy. By leveraging the random walks strategy, the RW-DO algorithm addresses the issue of premature convergence. This strategy enhances the diversity of solutions, thereby preventing the DO algorithm from becoming trapped in the local optima during the exploitation phase. To assess its performance, the RW-DO algorithm was compared with alternative algorithms by using the CEC 2020 and CEC 2019 function sets. Across all the test sets, the RW-DO algorithm consistently generateds more advantageous solutions. For the CEC 2020 function set, the RW-DO algorithm demonstrated superior performance compared with the DO algorithm from 3% to 11% in 5-dimensional problems. In 30-dimensional problems, the RW-DO algorithm exhibited superior performance compared to the DO algorithm from 14% to 46%. For the CEC 2020 function set in 50-dimensional problems, the RW-DO algorithm demonstrated superior performance compared to the DO algorithm from 11% to 188%. In the CEC 2019 function set, this ratio ranges from 3% to 38%. In engineering problems, the RW-DO algorithm also achieved superiority over the DO algorithm from 2% to 3%. Statistical analyses were performed to validate the superiority of RW-DO. The Kolmogorov-Smirnov normality test was used to select appropriate statistical tests for evaluating the performance of the algorithm based on the CEC 2020 and CEC 2019 function sets. Because the data set is not normally distributed, non-parametric tests such as the Wilcoxon signed-rank test and Kolmogorov-Smirnov for two sample tests were employed. These tests confirm that RW-DO yields distinct and superior solutions compared with the different data sets. Furthermore, the effectiveness of the RW-DO algorithm in solving real-world problems was demonstrated through its application to six engineering design problems. The experimental results highlight its competency in comparison to other algorithms. Overall, this research demonstrates the enhanced capabilities of the RW-DO algorithm in optimizing complex problems and its competitiveness when pitted against alternative methods.