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.
出处
《计算机仿真》
CSCD
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
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
Neural network
Integrated evaluation
Index optimization
Weight-spectrum
Contribution ratio