Random search with adaptive boundaries algorithm for obtaining better initial solutions. (July 2022)

• Record Type: Journal Article
• Authors: Öztaş, Gülin Zeynep; Erdem, Sabri
• Is Part Of: Advances in engineering software. Volume 169(2022)
• Publication Date: 2022-07
• Subjects: Metaheuristics -- Initial solution methods -- Adaptive random search -- Unconstrained optimization problems; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2022.103141
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: The proposed algorithm depends on updating upper and/or lower boundaries dynamically. This algorithm provides an adaptive initial solution which is better than pure random. The proposed algorithm converges to the optimal-like solutions in small dimension test cases. In large dimension cases, wide boundaries have been narrowed without trapping local optimums. RSAB is ready to be integrated with global optimum algorithm as the first step. Abstract: Conventional random search techniques take a lot of time to reach optimum-like solutions. Thus, random search techniques with advanced competencies play an essential role in algorithms. In this study, we develop an algorithm that provides an adaptive initial solution, to some extent reducing the diversity of randomness in the initialization of the algorithms for continuous unconstrained/bounded nonlinear optimization problems. The algorithm meets this expectation by narrowing search space adaptively without trapping into local optimums. It also escapes from eliminating accidentally global optimum in multi-modal problems. For this reason, we configure the proposed algorithm on the principle of updating given upper-lower boundaries dynamically. It is worth mentioning that this procedure does not add an additional burden to existing solution methods; on the contrary, it contributes to problem-solving in terms of time and efficiency. To show its performance, we have incorporated with most frequently used unconstrained/bounded benchmarks and compared them with the solutions in the literature. In conclusion, the proposed algorithm converges solutions quickly and is applicable for later usage in further studies.