多属性信息决策的改进无监督学习算法建模与应用

针对SOM网络无监督学习算法的单样本序列学习方式内存占用多的特点,采用Voronoi矢量原理改进权矢量迭代方式,使改进算法具有所有样本同时学习的能力,同时给出了算法的矢量映射误差测度和拓扑误差测度。然后根据改进算法建立了多属性信息决策的可视二维拓扑映射图模型,并对R&D项目中止决策进行了研究。计算结果表明,改进的无监督学习算法收敛速度快,基于拓扑映射图模型的多属性决策有效。

Multiple Attribute Information Decision Making Mode and Application Via Improved Fast Unsupervised Learning Algorithm

Aiming at deficiency of sample vector sequential training algorithm of Self-Organizing Map which uses much memory,the paper proposes a new training algorithm based on theory of Voronoi vector in which all weight vectors of samples can be adjusted in each training step,and also presents measure of vector mapping error and topographic error.Furthermore the paper set up a visualized topology map mode of multiple attribute decision making based on improved algorithm,and uses it to R&D project termination decision.The results show that the convergence of improved algorithm is faster and the decision making is valid based on topology map mode.