AUTOMATION CHOICE OF THE SOLUTIONS WITH PRECEDENTS USING BASED ON THE ALTERNATIVES FROM PARETO SET

Abstract

The paper introduces the approach to automation of Pareto set reduction and solutions selection using precedents. In multi-objective optimization process the problem to select Pareto-optimal solutions emerges very frequently. Such a choice is possible not based only on the expert’s preference scheme but with precedents using. They reflect the accumulated experience in decision making on the similar problems solving. The generating of these precedents based on the alternatives from Pareto set is suggested. The procedures for automation of the precedents base creating and precedents matching for new solutions analysis are offered also. In this approach the precedent is an aggregate of some obtained Pareto-optimal solution and their fitness rating for expert. Such fitness value may be assigned directly by him or determined by results of clusterization, so that the solutions from same cluster have been got the equal fitness values. Further the fitness value estimate of new solution of multi-objective optimization problem may be performed with machine learning methods using such as decision trees, artificial neural network without calling expert. The series of experiments for accuracy testing on fitness assessment of new solutions were conducted for a multi-objective optimization test problem. The high accuracy was achieved for a compact base of the precedents with a decision trees method using and neural networks using for relatively large base of the precedents were more effectively.