BP神经网络，relu做激活函数，求大神帮忙

Question

写了一个用relu实现BP神经网络的代码，但是发现relu使得ou输出都是0，但是还是不行，错误率高达60%，我是自学的，代码可能存在错误，下面是主要的代码，希望能够得到网上各位大神的帮忙，谢谢了

double* backward(double *delta)
{
    for (int i = 0; i < onum; i++)
    {
        deto[i] = delta[i];
        for (int j = 0; j < hnum; j++)
            wo[j][i] = wo[j][i] + learnrate*delta[i]*xh[j];
    }
    for (int i = 0; i < hnum; i++)
    {
        for (int j = 0; j < onum; j++)
            deth[i] += deto[j] * wo[i][j]*relubk(xh[i]);
        for (int j = 0; j < inum; j++)
            wi[j][i] += learnrate*deth[i]*xi[j];
    }
    for (int i = 0; i < inum; i++)
        for (int j = 0; j < hnum; j++)
            deti[i] += deth[j] * wi[i][j] * relubk(xi[i]);
    for(int i=0;i<onum;i++)
        obias[i]+= learnrate*delta[i];
    for (int i = 0; i < hnum; i++)
        hbias[i] += learnrate*deth[i];
    return deti;
}
我最后一级用了sigmoid 
double *dec(double *x, double *y, int l, double *det)
{
    double det2[N] = { 0 };
    for (int i = 0; i < l; i++)
    {
        xou[i] = 0;
        for (int j = 0; j < l; j++)
            xou[i] += x[j] * weight[j][i];
        xou[i] = 1.0 / (1.0 + exp(-xou[i] - bias[i]));
        //fl << "xou" << i << xou[i] << ' ';
    }
    fl << endl;
    for (int i = 0; i < l; i++)
    {
        det2[i] = (y[i] - xou[i])*(1-xou[i])*xou[i];
        fl << det2[i] << "," << y[i] << endl;

        for (int j = 0; j < l; j++)
        {
            weight[j][i] += learnrate*det2[i] * x[i];
        }

    }
    for (int i = 0; i < l; i++)
    {
        bias[i] += learnrate*det2[i];
    }
    for (int i = 0; i < l; i++)
    {
        det[i] = 0;
        for (int j = 0; j < l; j++)
            det[i] += relubk(x[i])*det2[j] * weight[i][j];
        fl << "det" << i << ":" << det[i] << ' ';
    }

Answer 1

问题解决否，同样的问题