The tesseract included in OpenAlpr has own pre-trained directory in runtime_data/runtime_data/ocr/tessdata - you can see that there are pre-trained models. Further, you can download particular pre-trained models, even Marrocan one, on https://github.com/tesseract-ocr/tessdata and that is all:) Respectively I do not know if the new pre-trained data file has to be included in Openalpr source code or are used automatically (I suppose that automatically). After that, no additional training is necessary, OpenAlpr should recognise Marrocan plate:)

You can see this workshop

http://asar.ieee.tn/