输出基于python的使用bp神经网络，含有双隐层的代码

当然可以，以下是一个基于Python实现的双隐层BP神经网络的代码示例：

import numpy as np

class NeuralNetwork:
    def __init__(self, input_layer_size, hidden_layer_size1, hidden_layer_size2, output_layer_size):
        self.input_layer_size = input_layer_size
        self.hidden_layer_size1 = hidden_layer_size1
        self.hidden_layer_size2 = hidden_layer_size2
        self.output_layer_size = output_layer_size

        self.weights1 = np.random.normal(0.0, pow(self.hidden_layer_size1, -0.5), (self.hidden_layer_size1, self.input_layer_size))
        self.biases1 = np.zeros((self.hidden_layer_size1, 1))
        self.weights2 = np.random.normal(0.0, pow(self.hidden_layer_size2, -0.5), (self.hidden_layer_size2, self.hidden_layer_size1))
        self.biases2 = np.zeros((self.hidden_layer_size2, 1))
        self.weights3 = np.random.normal(0.0, pow(self.output_layer_size, -0.5), (self.output_layer_size, self.hidden_layer_size2))
        self.biases3 = np.zeros((self.output_layer_size, 1))

    def sigmoid(self, x):
        return 1 / (1 + np.exp(-x))

    def softmax(self, x):
        return np.exp(x) / np.sum(np.exp(x), axis=0)

    def feedforward(self, inputs):
        inputs = inputs[:,np.newaxis]
        hidden_outputs1 = self.sigmoid(np.dot(self.weights1, inputs) + self.biases1)
        hidden_outputs2 = self.sigmoid(np.dot(self.weights2, hidden_outputs1) + self.biases2)
        final_outputs = self.softmax(np.dot(self.weights3, hidden_outputs2) + self.biases3)
        return final_outputs

    def train(self, inputs_array, targets_array, learning_rate):
        inputs_array = inputs_array.T
        targets_array = targets_array[:,np.newaxis].T

        hidden_outputs1 = self.sigmoid(np.dot(self.weights1, inputs_array) + self.biases1)
        hidden_outputs2 = self.sigmoid(np.dot(self.weights2, hidden_outputs1) + self.biases2)
        final_outputs = self.softmax(np.dot(self.weights3, hidden_outputs2) + self.biases3)

        output_errors = targets_array - final_outputs
        hidden_errors2 = np.dot(self.weights3.T, output_errors) * hidden_outputs2 * (1 - hidden_outputs2)
        hidden_errors1 = np.dot(self.weights2.T, hidden_errors2) * hidden_outputs1 * (1 - hidden_outputs1)

        self.weights3 += learning_rate * np.dot((output_errors * final_outputs * (1 - final_outputs)), hidden_outputs2.T)
        self.biases3 += learning_rate * output_errors * final_outputs * (1 - final_outputs)
        self.weights2 += learning_rate * np.dot(hidden_errors2, hidden_outputs1.T)
        self.biases2 += learning_rate * hidden_errors2
        self.weights1 += learning_rate * np.dot(hidden_errors1, inputs_array.T)
        self.biases1 += learning_rate * hidden_errors1

备注：以上代码仅供参考，可能有改进的空间，需要根据具体任务进行调整。