从生物学和逼近论出发 ,将任意一组正交基函数作为三层前向神经网络各隐含神经元的活跃函数 ,再以其加权和作为网络的输出特性 ,构成一种新型的神经网络模型 ;从理论上解决了单输入神经网络隐含层数及隐含单元个数难以确定的问题。仿真实验表明 ,该网络具有优良的逼近任意非线性对象的特性 ,且收敛速度远远快于 BP网络。 Based on biology and approximation theory, an arbitrary set of orthogonal basis functions are taken as the active function of each implicit neuron in the three-layer feedforward neural network, and then the weighted sum is used as the output characteristic of the network to form a new type of neural network Model; solves the problem that the number of hidden layers and the number of hidden units in a single-input neural network can not be determined theoretically. Simulation results show that this network has excellent characteristics of approximating any non-linear object, and the convergence speed is much faster than that of BP network.