神经网络的结构设计一直是研究者关注的基本问题.提出了一种构造神经网络的方法———EBNN.该方法采用熵作为第一层神经元的选择准则,在选择第一层神经元的同时可以得到隐层和输出层的神经元,并对神经网络参数进行合理的初始化.该方法不仅能有效地建立神经网络的结构,还提供了信息性增量学习的可能性. The structural design of neural network has always been the basic problem of researchers, and a method of constructing neural network (EBNN) is proposed.This method uses entropy as the selection criterion of the first neurons, At the same time, the neurons in hidden layer and output layer can be obtained, and the parameters of neural network can be reasonably initialized. This method can not only effectively construct the neural network structure, but also provide the possibility of informative incremental learning.