Performance improvement of learning feedforward control