Black-box Multi-objective Optimization Evaluation System Research And Implementation

Optimization design is a disign method to choose the best program from a variety of programs, which contains two aspects:the first is to get a variety of programs, and then is to choose from a variety of programs, the former belongs to multiple objective decision making, and the later belongs to multiple attribute decision making. Usually, the practical optimization problems which the designers face to contain a number of conflicting and commensurability decision-making targets. Due to low quality of the traditional design methods, designers cannot get satisfactory design programs and the efficiency of calculation is not high. Therefore, optimization design technology has been widely applied in different engineering fields in recent years, and the research of optimization design has been receiving more and more attention.This paper studied the "black box" multi-objectives optimization evaluation problem which has no mathematical model, on this basis, the paper developed an optimization evaluation system based on the commercial software of Matlab, to solve the black-box multi-objective optimization evaluation problems which are complex to solve, complicated to deal with, and difficult to applicate. The optimization evaluation process of the system is described as follows:select the appropriate experiment design method and the corresponding experiment design table, and get the representative input and output samples by experiment; choose the samples which contain all patterns to train artifical neural network and get network model between parameter variables and targets; use the improved elitist fast non-dominated sorting genetic algorithm (a variant of NSGA-Ⅱ) to optimizate multi-objective parameters, and deal with constraint conditions at the same time, and get the Pareto optimal solutions which satisfy the constraint; finally the system provides four kinds of optimal solutions evaluation methods:based on intuitionistic fuzzy and entropy TOPSIS evaluation method, based on TOPSIS robustness evaluation method, fuzzy closeness evaluation method and visualization radar graphical evaluation method to evaluate the Pareto optimal solutions obtained by optimization, and then determines the optimal solution from them. The system also introduces external interface technology to receive third-party data files, increasing the practicality of optimization evaluation system.The research of paper shows that the development of multi-objective optimization assessment system can get better optimization design results, and improve the disadvantage of the previous optimization algorithm which can only get the Pareto optimal solutions but cannot get the optimal solution, at the same time improve the lack of calculation speed and convergence precision, and provide an effective solution for lack of mathematical model of the "black box" multi-objective optimization problem.