Particle swarm optimization for dynamically changing environments with particular focus on scalability and switching cost

Change is an inescapable aspect of natural and artificial systems, and adaptation is central to their resilience. Optimization problems are no exception to this maxim. Indeed, viability of businesses depends heavily on their effectiveness in responding to a change in the myriad of optimization problems they entail. Changes in optimization problems usually are result of change in the objective function and/or number of variables and/or constraints. Such optimization problems are denoted as dynamic optimization problems (DOPs) in the literature. Despite the large body of literature on DOPs and algorithms in this domain, there are still noticeable gaps between real-world DOPs and academic research. The first objective of this thesis is investigating DOPs to identify any class of DOPs or any DOPs' characteristics that are common in practical situation but have not been studied by the researchers. In this thesis, two important gaps are identified, namely considering switching cost in DOPs and large-scale DOPs. Both are common in many real-world dynamic problem but a few research investigated them in the past. In an attempt to bridge these gaps, this thesis makes the following contributions: First, this thesis considers the impact of cost for changing solutions after environmental changes. In fact, changing solutions in real-world problems is costly. Furthermore, larger changes have higher cost and need more resources such as time, human resources and energy. Thus, lack of switching cost consideration in most previous algorithms makes them unsuitable for many of real-world DOPs. In this thesis, different scenarios of DOPs with switching cost are investigated, their challenges are identified, and the performance of the state-of-the-art methods are investigated for solving them. Contributions include developing a novel robust optimization over time (ROOT) framework, a novel adaptive method for maximizing efficiency by changing or keeping solutions after environmental changes, and a novel multi-objective and time-linkage based method for minimizing switching cost. Second, this thesis investigates large-scale DOPs. Up to now, little attention has been given to the scalability of DOPs. Indeed, the dimension of typical DOPs studied in the literature hardly exceeds twenty. In this thesis, the challenges of large-scale DOPs are studied, then the efficiency of the current methods are investigated for solving them. Moreover, this thesis proposes a novel cooperative coevolution algorithm based on a multi-population approach which benefits from a new resource allocation method for DOPs with high-dimensional search space. All the proposed methods in this thesis use particle swarm optimization as the core optimizer embedded in a multi-population framework. The performance of the proposed methods are compared with state-of-the-art methods on a wide range of problem instances generated by the state-of-the-art and the proposed DOP benchmarks. The comparison results indicate the superiority of the proposed methods