Methods of enhancing the MOO CEM algorithm

Trankle, Veronique (2021-03)

Thesis (MEng)--Stellenbosch University, 2021.


ENGLISH ABSTRACT: Optimisation refers to solving a problem as accurately as possible while making the most effective use of the available resources. Whilesome problems may have a single best solution, problems with more than one objective often do not have a single optimal solution. These types of problems are known as multi-objective problems. Improving the performance of multi-objective optimisation algorithms, both interms of accuracy and execution time, is an active topic of research,with great emphasis being placed on computational efficiency. The MOO CEM algorithm was developed as an alternative algorithmto solve multi-objective optimisation problems accurately and efficiently. However, the algorithm has a number of limitations. This study explores methods to improve and enhance the existing MOOCEM algorithm. These areas of improvement include: improving the sampling method of the algorithm and enhancing the algorithm by adding functionality to solve constrained problems and problems with more than two objective functions. Following a thorough literature study of appropriate techniques, two methods of sampling improvement were identified: the Beta distribution and the use of covariance of the decision variables. In terms of solving constrained problems, two methods were evaluated: the elimination method and the dynamic penalty method. The ENS-SSalgorithm was selected as the ranking and selection method, enabling the algorithm to sort (and thereby solve) problems with more than three objectives.The proposed improved and enhanced algorithms were tested individually on a number of benchmark problems. Pareto-compliant performance indicators were used to evaluate the performance of the algorithms and, where possible, statistical tests were conducted to compare the performances to that of the original MOO CEM algorithm. It was observed that the use of the Beta distribution improved the performance of the algorithm, while using covariance did not.Adding the constraint and multi-dimensional ranking functionality yielded positive results. Following the outcome of the tests, a final algorithm was proposed,incorporating the successful elements of the study. The final algorithm is relatively simple to implement and improves and expands the functionality of the original MOO CEM algorithm.

AFRIKAANSE OPSOMMING: Raadpleeg teks vir opsomming

Please refer to this item in SUNScholar by using the following persistent URL:
This item appears in the following collections: