A Bio-inspired algorithm to solve Dynamic Multi-Objective Optimization Problems
Martínez Peñaloza, María Guadalupe · 2018
Many different applications in engineering, science and industry have a considerable degree of complexity, and sometimes this complexity may be based on the presence of multiple conflicting objective functions, which must be simultaneously optimized. These kinds of problems are the so-called multi-objective optimization problems (MOPs). However, in everyday life, most optimization problems are not static in nature and usually have at least one objective that can change over time. In recent years, MOPs in dynamic environments have attracted some research efforts. However, most research focuses on either static multi-objective optimization or dynamic single-objective optimization. Therefore, not much research has been done on Dynamic Multi-Objective Optimization (DMO). Evolutionary algorithms anMany different applications in engineering, science and industry have a considerable degree of complexity, and sometimes this complexity may be based on the presence of multiple conflicting objective functions, which must be simultaneously optimized. These kinds of problems are the so-called multi-objective optimization problems (MOPs). However, in everyday life, most optimization problems are not static in nature and usually have at least one objective that can change over time. In recent years, MOPs in dynamic environments have attracted some research efforts. However, most research focuses on either static multi-objective optimization or dynamic single-objective optimization. Therefore, not much research has been done on Dynamic Multi-Objective Optimization (DMO). Evolutionary algorithms and Artificial Immune System (AIS) have been popular to solve dynamic single objective optimization problems. Nevertheless, such combination has been scarcely explored when solving DMOPs.d Artificial Immune System (AIS) have been popular to solve dynamic single objective optimization problems. Nevertheless, such combination has been scarcely explored when solving DMOPs. On the other hand, a few Differential Evolution(DE)-based algorithms have been proposed. In this thesis, two Differential Evolution-based algorithms to solve dynamic multi-objective optimization problems (DMOPs) are proposed. The novelty of these algorithms with respect to other approaches is the fact that the algorithms take advantage of DE and AIS to track the changes in the environment and respond quickly when a change is detected. Three main issues of the algorithms are explored: (1) the general performance of both algorithms in comparison with other well-known algorithms, (2) their sensitivity to different change severities and frequencies, and (3) the role of their change reaction mechanism based on an immune response. For such purpose, different performance metrics, four unary and one binary, are computed in a comparison against other state-of-the-art dynamic multi-objective evolutionary algorithms (DMOEAs) when solving a novel suite of test problems. The statistically validated results indicate that the proposed approaches are robust to change frequency and severity variations and can track the environmental changes finding a good distribution of solutions.