High-dimensional multi-target oriented multi-population mixing evolution method

Abstract

The invention provides a high-dimensional multi-target oriented multi-population mixing evolution method. A fixed direction matrix covering a whole searching space is generated by means of a sine function, and high-dimensional multi-target optimization is turned into single-target optimization in each fixed direction; according to the concepts of leading bees and following bees in optimizing of an artificial bee colony, a multi-population mechanism is set, a following population is set for each direction, the optimal solutions of all directions are selected to constitute a leading population, and the leading population guides evolution searching of the following populations in all directions; a mixed evolution strategy is put forward, the convergence capacity in the fixed directions is enhanced by means of direction angle difference operators which are put forward, and meanwhile, local searching capacity is enhanced by means of SBX operators; an elitism strategy based on novel fuzzy domination is put forward to maintain the scale of an external archive set. According to the method, convergence and distributivity of the optimal solutions of high-dimensional multi-target optimization can be effectively improved, and the solving effect is not influenced by the number of targets.

Description

technical field [0001] The invention relates to an optimization method, in particular to an optimization method for high-dimensional multi-objectives. Background technique [0002] In multi-objective optimization, the improvement of one sub-objective may lead to the performance degradation of one or several other sub-objectives. In order to achieve the optimization of the overall objective, it is usually necessary to comprehensively consider the conflicting sub-objectives, that is, for each sub-objective compromise. Therefore, unlike single-objective optimization problems, multi-objective optimization does not have an absolute or unique best solution, but a set of optimal solutions composed of many Pareto optimal solutions. When the number of targets increases to 4 or more (called high-dimensional multi-target), the performance of these Pareto sorting methods will be greatly reduced, because as the number of targets increases, the individuals in the population do not domina...

AI Technical Summary

Problems solved by technology

The first type is the weight coefficient method. This type of method transforms the high-dimensional multi-objective optimization problem into a single-objective optimization problem by designing a set of weight coefficients. The weight coefficient introduced is equivalent to the introduction of preference information, but the information is not accurate and comprehensive. , will cause the optimization result to deviate from the true Pareto front; the second type is the dimensionality reduction method, which compares the dominance relationship between the solutions under the condition of ignoring or excluding certain targets through a certain strategy, but the target The reduction of will cause the loss of information, so this type of method is optimized under the minimum allowable error, and the application effect will be affected by the increase in the dimension of the Pareto front; the third type is based on the loose domination method, the idea of ​​​​this type of method is in After reducing or enlarging the objective function value of the individual according to a certain ratio, and then comparing it with other individuals, it is actually a loose improvement of Pareto domination. The parameters in this method are difficult to determine, and the real objective function of the individual is changed. value, will cause the solution method to fail to converge to the real Pareto front; the fourth category is the decomposition method, which converts the high-dimensional multi-objective optimization problem into multiple single-objective optimization problems through a set of weight vectors and solves them simultaneously. Good results have been achieved on the optimization problem, but the transformation process does not consider the influence of the value range of each target, and the evolutionary operation is not perfect enough, resulting in insufficient convergence and easy to fall into local optimum

Images