基于神经网络的指标体系优化方法  

Neural Network-based Optimization Method of Index System

作　　者：陈海英[1] 郭巧[1] 徐力[1] 

机构地区：[1]北京理工大学机器人研究中心,北京100081

出　　处：《计算机仿真》2004年第7期107-109,118,共4页Computer Simulation

基　　金：国家自然科学基金资助项目(60171018)4-3.2574-2.2905-0.0064

摘　　要：针对现有指标优化方法在非线性系统应用中的不足,该文提出综合评价指标体系的神经网络指标优化方法。即以核心指标作为网络输出,其他因素指标作为网络输入,建立前向神经网络模型,通过网络刻画出输入和输出之间的相关性。进而选择与输出相关程度大的输入指标作为优化指标。这种方法不需先验假设、建模的过程简化、可以避免主观因素对变量选取的干扰,同时精度也有很大提高。In view of the deficiency of existing index optimization methods in non-linear system application, this paper presentsa neural network-based optimization method for the index system of integrated evaluation. It takes the core index as output of network, other indexes as inputs of network and establishes forward neural network model. Correlation between input and output can be depicted by network. And it picks out some input indexes which have high-correlation with output as optimization indexes. This method doesn't need a priori hypothesis, the process of modeling is simple, and can avoid disturbance of subjective factors to the selection of variables. At the same time the precision is also improved.

关 键 词：神经网络 指标优化 贡献率 综合评价 非线性系统 权值谱 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]