You need to first decide about the factors that affect the discharge. You can do it by referring to previous papers.
Then create your database in which the target values (discharge and peak time) are attributed to input variables. When you let the network know these factor as outputs, it will predict them using input variables.
@Mohammad Mehrabi thank you. IT means i need to input rainfall as input data and discharge and time in target value to get predicted discharge. Is this a right way ?
I completely agree with @Mohammad Mehrabi. You can make your dataset with rainfall events as input and discharge along with time as output. Thereafter, you can use BR or LM learning method for training.
However, you should once verify that whether your input will be 8×476 or 476×8, as per your data being related with each other.