本文提出了一类自我完善(self-improvernent)的神经元网络模型,包括一个学习过程和联想过程(retrieval procedure)。根据输入模式内在的规律性,学习算法通过统计规则确定神经元间的连接权。这种学习算法可以用一个多层次逻辑(logic)神经元网络实现。自我完善是通过联想过程实现的。这类神经元网络的特点包括无须预先确定训练模式,学习算法只需一次扫描,以及收敛速度快。计算机实验结果证明了神经元网络的自我完善能力。 This paper presents a self-improvernent neural network model that includes a learning process and a retrieval procedure. According to the inherent regularity of the input mode, the learning algorithm determines the connection rights between neurons by statistical rules. This learning algorithm can be implemented using a multi-level neural network. Self-improvement is achieved through the association process. Features of such neuronal networks include no need to predetermine training patterns, a learning algorithm requires only one scan, and fast convergence. Computer experimental results prove that the neural network self-improvement ability.