Step length progressive reduction-based evolutionary strategy optimization algorithm

Abstract

The invention discloses a step length progressive reduction-based evolutionary strategy optimization algorithm, which is an intelligent optimization algorithm originating from population competition, reproduction and evolution, and belongs to the technical field of optimization methods. According to the algorithm, a solution of a to-be-optimized problem is represented with an individual in a population, the initial evolutionary intensity is determined by the size of a feasible region of the problem, and an initial evolutionary direction is set to be isotropic. With continuous evolution of the population, the evolutionary intensity and the evolutionary direction of the population gradually adapt to a current environmental pressure, so that a more excellent population can be reproduced by the next generation. The quality of the individual in the population is determined by a fitness value and the environmental pressure is determined by a local form of a fitness function. For a nonlinear optimization problem or convex problem with a large amount of local extreme points, the algorithm has very high solving efficiency, can be converged with any precision and is simple to realize, very good in global search capability and high in convergence speed.

Description

technical field [0001] The invention relates to an evolution strategy optimization algorithm with decreasing step size, and belongs to the technical field of intelligent optimization algorithms. Background technique [0002] At present, the widely used classic optimization algorithms, such as: linear programming method, negative gradient method, Newton method and interior point method, etc., mainly apply to a class of linear, continuous and convex optimization problems. The main problem of applying this kind of optimization algorithm is that most engineering optimization problems have the characteristics of nonlinearity, non-convexity and multiple local extremum points. Although it has been theoretically proved that for convex optimization problems, such algorithms can quickly converge to the global optimal solution of the problem; but for non-convex and multi-local optimization problems, such algorithms often can only guarantee random convergence To a certain local extreme...

AI Technical Summary

Problems solved by technology

[0003] Aiming at the difficult implementation problems faced by the current mainstream evolutionary strategies, such as: the step size and covariance matrix update strategy are complicated, and the specific implementation and its effect depend on the selection of many parameters, a new intelligent optimization algorithm is improved and invented, namely Evolutionary strategy with decreasing step size

Images