You can start working on your problem by using nnstart. Set aside some samples of your historical data for validating the trained network and based on the validation results you can decide on creating a nerual network by yourself.
If your goal is to predict series y(t) given d past values of y(t) and another series x(t) then go for NARX or if your goal is to predict series y(t) given d past values of y(t) then go for NAR else your goal is to predict series y(t) given d past values of series x(t) then go for Nonlinear Input-Ouput.
The following link might help you in the process: Neural Network Time-Series Prediction and Modeling
