Genetic Algorithms for Solving Problems with Dynamic Active Variables and Dynamic Boundaries

Optimisation is a challenging research topic that relates to most real-life applications, such as transportation, management and industry. It aids in the search for optimal solutions to a research problem. Classifications of optimisation problems can be based on different perspectives, one of which is change over time whereby when an optimisation problem does not change it is called static, otherwise dynamic. Accordingly, when at least one part of an optimisation problem changes over time, it is called a Dynamic Optimisation Problem (DOP). Over the last few decades, fitness functions and constraints have been the most common factors of dynamism in DOPs. However, there is a strong need to develop and consider others, such as changes in active variables and boundaries which have not been well studied. This research addresses significant gaps in the literature, so one of its major contributions is developing and investigating new DOPs which take into consideration dynamism in both their active variables and the boundaries of variables, and formulates and designs general frameworks for them. The first DOP is one with unknown active variables (DOPUAV), the activity of which change over time. The second and third DOPs are ones in which the boundaries of the variables that change over time are known and unknown to the algorithm during the solving process: DOPKB and DOPUB, respectively. The fourth DOP is one with both unknown active variables and boundaries (DOPUAVB), that is, it consists of a DOPUAV and DOPUB.Also, the contributions include proposing two main approaches for solving these problems, i.e., problem information-based and population information-based. The former solves a problem based on the information known about the problem, such as its number of variables and their boundaries, while the latter uses information extracted from current solutions. To demonstrate the consistent performances of all the best proposed algorithms, they are compared with two of the most comprehensively studied algorithms for solving DOPs, namely, the Hyper Mutation GA (HyperM) and Random Immigration GA (RIGA), as well as the Simple GA (SGA). The comparisons are based on a modified version of one of the most commonly used measures of optimality, the average of best-of-generation (ABOG), as well as the average numbers of feasible changes (AFCh) and average feasibility percentages (AFP). In this thesis, the results show that the population information-based approaches are the most successful outperforming the three commonly used algorithms when solving the proposed DOPs