Multi-strategy coevolving aging particle optimization.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Black-box optimization problems are prevalent in science and engineering.
• Existing population-based algorithms face challenges in efficiency and robustness.
• Novel approaches are needed to enhance optimization performance.

Purpose of the Study:

• To introduce a novel population-based algorithm, Multi-Strategy Coevolving Aging Particles (MS-CAP).
• To evaluate the effectiveness and versatility of MS-CAP on benchmark problems and a real-world application.
• To compare MS-CAP's performance against state-of-the-art optimization algorithms.

Main Methods:

• MS-CAP combines two complementary algorithmic components in a memetic approach.
• Stage 1: Independent particle perturbation with decaying radius and increasing attraction to the best solution.
• Stage 2: Multi-strategy mutation and recombination inspired by Differential Evolution.

Main Results:

• MS-CAP was tested on two black-box optimization benchmarks at various dimensionalities.
• The algorithm was applied to train a Feedforward Neural Network for an 8-link robot manipulator.
• Numerical results indicate MS-CAP outperforms state-of-the-art algorithms on a majority of tested problems.

Conclusions:

• MS-CAP is a robust and versatile optimizer for black-box problems.
• The hybrid approach effectively balances exploration and exploitation.
• MS-CAP shows significant potential for complex optimization tasks in diverse fields.