Improved Particle Swarm Optimization Based on Gradient Descent Method

Abstract

Due to the lack of effective guidance on particle's speed and precocity in standard particle swarm optimization, a particle swarm optimization on the basis of gradient information and time-varying acceleration coefficient (TVAC), namely gradient descent particle swarm optimization (GDPSO), is proposed. By combining the direct method and the indirect method to solve the unconstrained optimization problem, the gradient information is used to modify the velocity term, guide the particle to conduct local efficient search, and improve the global explore ability of the algorithm through time-varying acceleration coefficient strategy. On the basis of simulation experiment and comparison with other algorithms, the proposed particle swarm optimization enjoys a fast convergence speed and is not easy to get trapped into local optimal, with excellent ability to solve complex multi-modal problems.