Investigation of Optimization Algorithms
Location
Ada, Ohio
Start Date
3-12-2024 12:00 AM
End Date
3-12-2024 12:00 AM
Description
My research investigates the performance of five optimization algorithms, Evolutionary Strategies (ES), Genetic Algorithm (GA), Differential Evolution (DE), Sequential Least Squares Programming (SLSQP), and Trust-Region Constrained (Trust-Constr), for value function optimization in the PI-NSGA-II package. I evaluated these algorithms on 100 value functions across different Pareto-front types (concave, convex, and linear) and varying numbers of decision-making points (η). My results show that GA and DE are unsuitable for this type of optimization, consistently underperforming and often resulting in negative epsilon values. In contrast, SLSQP, ES, and Trust-Constr demonstrated acceptable performance, with Trust-Constr offering a balance between accuracy and efficiency. After considering factors such as runtime, number of function evaluations, and final ε, I selected Trust-Constr as the default method for the PI-NSGA-II package.
Recommended Citation
Spinner, Elijah K., "Investigation of Optimization Algorithms" (2024). College of Engineering Student Research Colloquium. 4.
https://digitalcommons.onu.edu/eng_student_research_colloquium/2024/Presentations/4
Investigation of Optimization Algorithms
Ada, Ohio
My research investigates the performance of five optimization algorithms, Evolutionary Strategies (ES), Genetic Algorithm (GA), Differential Evolution (DE), Sequential Least Squares Programming (SLSQP), and Trust-Region Constrained (Trust-Constr), for value function optimization in the PI-NSGA-II package. I evaluated these algorithms on 100 value functions across different Pareto-front types (concave, convex, and linear) and varying numbers of decision-making points (η). My results show that GA and DE are unsuitable for this type of optimization, consistently underperforming and often resulting in negative epsilon values. In contrast, SLSQP, ES, and Trust-Constr demonstrated acceptable performance, with Trust-Constr offering a balance between accuracy and efficiency. After considering factors such as runtime, number of function evaluations, and final ε, I selected Trust-Constr as the default method for the PI-NSGA-II package.