本文介绍一个利用人工神经网络进行分型和自动知识获取的新一代精神分裂症诊断系统。阐述了训练样本的统计方法，神经网络的模糊与清晰性相结合的分型方法以及动态自学习方法，在输出规则中引入了黑洞束缚的概念。 This article presents a new generation of schizophrenia diagnostics that uses artificial neural networks for typing and automated knowledge acquisition. The statistical methods of training samples, the combination of fuzzy and sharpness of neural networks and the dynamic self-learning method are introduced. The concept of black hole constraint is introduced in the output rules.