说明：  编程实现三层前馈神经网络 BP 学习算法，该程序具有以下功能： 1. 允许选择各层节点数； 2. 允许选用不同的学习率； 3. 能对权值进行初始化，初始化用[-1，1]区间的随机数； 4. 允许选用单极性或双极性两种不同的 S 型转移函数。 5. 要求有建立网络、训练网络和测试网络的功能。 6. 采用上述 BP 网络逼近非线性对象： 采样时间取 1ms，输入信号为 7. 没有动量因子时的逼近效果和逼近误差。 8. 考虑加入动量因子 ?（? 的值自定）的逼近效果和逼近误差。
(The three level feedforward neural network BP learning algorithm is implemented by programming. The program has the following functions: 1. the number of nodes in each layer is allowed. 2. the selection of different learning rates is allowed. 3. initialization of weights and initialization of random numbers in [-1 and 1] intervals; 4. two types of S transfer functions are allowed to be unipolar or bipolar. 5., it requires the functions of establishing network, training network and testing network. 6. the nonlinear object is approximated by the above BP network. The sampling time is 1ms, and the input signal is 7. the approximation effect and approximation error when there is no momentum factor. 8. consider the approximation effect and approximation error of adding momentum factor.)