A multiagent genetic algorithm for global numerical optimization

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Artificial Intelligence

Background:

• Global numerical optimization is a challenging problem with broad applications.
• Existing algorithms may struggle with high-dimensional problems and scalability.
• Multiagent systems offer a framework for distributed problem-solving.

Purpose of the Study:

• To introduce a novel algorithm, the Multiagent Genetic Algorithm (MAGA), by integrating multiagent systems and genetic algorithms.
• To evaluate the performance and scalability of MAGA for global numerical optimization.
• To demonstrate the applicability of MAGA to practical problems like linear system approximation.

Main Methods:

• MAGA represents candidate solutions as agents in a lattice-like environment.
• Agents interact with neighbors through competition and cooperation, utilizing knowledge.
• The algorithm aims to minimize the objective function value through agent interactions.

Main Results:

• MAGA demonstrated good performance on ten benchmark functions.
• The algorithm exhibits excellent scalability, maintaining effectiveness with problem dimensions up to 10,000.
• MAGA successfully found high-quality solutions for high-dimensional problems at a low computational cost.
• Application to linear system approximation yielded satisfactory results.

Conclusions:

• MAGA is a competent algorithm for solving high-dimensional global numerical optimization problems.
• The novel integration of multiagent systems and genetic algorithms provides significant advantages in scalability and performance.
• This research pioneers the optimization of functions with 10,000 dimensions using evolutionary approaches.