请生成一个输入层4个节点，第一个隐藏层5个节点，第二个隐藏层4个节点，输出层1个节点，激活函数为ReLU,全连接的BP神经网络matlab训练代码

好的，以下是生成您所需神经网络的MATLAB训练代码：

% 设置网络结构及参数
input_nodes = 4;
hidden_1_nodes = 5;
hidden_2_nodes = 4;
output_nodes = 1;
activation_function = 'ReLU';

% 随机初始化网络参数
w1 = randn(hidden_1_nodes, input_nodes);
b1 = randn(hidden_1_nodes, 1);
w2 = randn(hidden_2_nodes, hidden_1_nodes);
b2 = randn(hidden_2_nodes, 1);
w3 = randn(output_nodes, hidden_2_nodes);
b3 = randn(output_nodes, 1);

% 设置学习率和训练轮数
learning_rate = 0.1;
epoch = 10000;

% 加载训练数据集，并将目标值转换为矩阵形式
load training_data.mat
Y = Y';

% 开始训练
for i = 1:epoch
% 前向传播
Z1 = w1*X + b1;
A1 = activation(Z1, activation_function);
Z2 = w2*A1 + b2;
A2 = activation(Z2, activation_function);
Z3 = w3*A2 + b3;
A3 = Z3;

% 计算代价函数并输出当前代价
cost = sum((A3-Y).^2) / size(Y, 2);
fprintf('Epoch %d, cost: %.6f\n', i, cost);

% 反向传播
delta3 = A3-Y;
delta2 = w3'*delta3 .* derivative_activation(Z2, activation_function);
delta1 = w2'*delta2 .* derivative_activation(Z1, activation_function);

% 更新参数
w3 = w3 - learning_rate*delta3*A2';
b3 = b3 - learning_rate*sum(delta3, 2);
w2 = w2 - learning_rate*delta2*A1';
b2 = b2 - learning_rate*sum(delta2, 2);
w1 = w1 - learning_rate*delta1*X';
b1 = b1 - learning_rate*sum(delta1, 2);
end

% 定义激活函数及其导数函数
function output = activation(x, activation_function)
if strcmp(activation_function, 'ReLU')
output = max(0, x);
elseif strcmp(activation_function, 'sigmoid')
output = 1 ./ (1+exp(-x));
elseif strcmp(activation_function, 'tanh')
output = tanh(x);
end
end

function output = derivative_activation(x, activation_function)
if strcmp(activation_function, 'ReLU')
output = sign(x>0);
elseif strcmp(activation_function, 'sigmoid')
output = activation(x, activation_function) .* (1 - activation(x, activation_function));
elseif strcmp(activation_function, 'tanh')
output = 1 - (activation(x, activation_function)).^2;
end
end

% 保存最终参数
save('trained_weights.mat', 'w1', 'b1', 'w2', 'b2', 'w3', 'b3');