similar time, worse prediction

Today, I build an ANN model in r exactly the same as what it is in matlab. But the prediction result is much worse than that in matlab.

Summary:

I build an ANN model with same hidden layers and neurons, same training function (scg), same initial weights and biases (0.1), same training (85%) and testing (15%) data.
The result shows that in matlab, the time for 500 iterations is 2.68 s, while in r, the time for 500 iterations is 2.53 s. It seems that r is even a little better. (I was wrong because I did not control the same training function.)
However, the prediction result of the ANN model in r is very poor, which is shown below.

For most data, the prediction result is about 0 no matter what it is originally. For now, I do not know why it is like that.

Tomorrow, I will try to fix the problem.