Retraining nlarx using init()

I have identified a Neural Network by means of nlarx function.

I would like to retrain this network starting from the weights and bias obtained from the first training.

I considered the function init() taken from the documentation to reinitialize the new training with the same network architecture, weights and bias of the first network (model_narx).

net = model_narx.Nonlinearity.Network;
net = init(net);

However if I perform a new training it seems that it takes into account the desired architecture but neglects the weights and bias.

Indeed the performance does not start from the optimal value of the first training even when using the same data set used to train the network for the first time.

Thank you for the support.