超声波铁轨检测数据.

Question

需要一份超声波检测铁轨损伤的数据，能用elman神经网络算法仿真出来最好。（可加钱）

Answer 1

python
import numpy as np
import matplotlib.pyplot as plt
 
# 激活函数
def sigmoid(x): 
    return 1 / (1 + np.exp(-x))
 
# 构建Elman神经网络
def elman_net(X, T, H_nodes, O_nodes, eta, epochs):
    # 输入层、隐藏层、输出层的权重矩阵和阈值向量
    W_input = np.random.rand(X.shape[1], H_nodes) 
    W_hidden = np.random.rand(H_nodes, H_nodes)
    W_output = np.random.rand(H_nodes, O_nodes)
    bias_input = np.random.rand(H_nodes) 
    bias_hidden = np.random.rand(H_nodes) 
    bias_output = np.random.rand(O_nodes)
 
    # 训练神经网络  
    for i in range(epochs): 
        # 信号前向传播  
        hidden_layer = sigmoid(np.dot(X, W_input) + bias_input)  
        x_hidden = np.hstack((np.zeros((hidden_layer.shape[0], X.shape[1])), hidden_layer))  
        hidden_layer = sigmoid(np.dot(x_hidden[:, :-H_nodes], W_hidden) + bias_hidden)  
        output = sigmoid(np.dot(hidden_layer, W_output) + bias_output)
 
        # 误差反向传播
        E = T - output
        mse = E ** 2
        bias_output -= eta * mse.mean(axis=0)
        W_output -= eta * np.dot(hidden_layer.T, E) / E.shape[0] 
        E = (E * W_output).dot(hidden_layer * (1 - hidden_layer)) 
        bias_hidden -= eta * E.mean(axis=0)
        W_hidden -= eta * np.dot(x_hidden[:, :-H_nodes].T, E) / E.shape[0]
        E = (E * W_hidden).dot(x_hidden[:, :-H_nodes] * (1 - x_hidden[:, :-H_nodes])) 
        bias_input -= eta * E.mean(axis=0)
        W_input -= eta * np.dot(X.T, E) / E.shape[0]
    return W_input, W_hidden, W_output, bias_input, bias_hidden, bias_output