You can try d=1 and check the stationarity by Augmented Dickey Fuller test, in case there is still nonstationarity you can try d=2.

You can use auto.arima function from 'forecast' package for R.

Alternatively, if you have many observations, you can try out-of-sample comparison of alternative models with different values of d.

To compare alternative models, you can use the instructions described here:

Chapter Forecast Error Measures: Critical Review and Practical Recommendations

You can difference the series once and test it for stationarity. If stationary at this point d=1 obtain the ACF and the PACF their cut-off points give q and p respectively. If not stationary still, difference it again, test for stationarity, if stationary d=2, obtain the ACF and the PACF, their cut-off points give q and p respectively.

You can try some grid search method to determine optimal parameter d.

Something like this -> https://machinelearningmastery.com/grid-search-arima-hyperparameters-with-python/