线性判别函数理论是线性分类器的分析基础,并不适合非线性分类器。本文把非线性激励函数视为隶属度函数,将非线性神经元及多层感知器分类行为的分析建筑在模糊集理论基础上,提出模糊线性判别函数与模糊判别边界、模糊分类等概念。并引出将隐层初始权向量均匀分布在权空间超球面上的初始化方法,明显提高了多层感知器的收敛性能,并提出了一种在多层感知器的类空间中构造最优超平面的简易新方法。 Linear discriminant function theory is the basis of linear classifier analysis, not suitable for nonlinear classifier. In this paper, the non-linear excitation function is regarded as a membership function. Based on the analysis of the classification of nonlinear neurons and multi-layer perceptrons, the concepts of fuzzy linear discriminant function, fuzzy discriminant boundary and fuzzy classification are proposed. The initialization method of uniformly distributing the hidden layer initial weight vector on the hypersphere of the weight space is derived, which obviously improves the convergence performance of multi-layer perceptrons and proposes a method of constructing optimal hyperplane in the class space of multi-layer perceptrons Simple and easy way