there is a great variety of pv modules out there. Their reflection properties depend mostly on the front glass layer, which may be heavily textured. One can measure the BRDF of such modules (given that a representative area is covered, since typical module surfaces are spatially nonuniform) and generate a data-driven model from that. This approach was applied e.g. in a reflection study .
If the exact surface structure is known, one can also develop a “layered” model for the perturbated glass surface, the conductors and the (hardly reflecting) background of cells and back-sheet.
Andreas Noback’s Ph.D. work on modelling glass mosaic may come surprisingly close to that problem . Unfortunately the thesis is in German and not published yet (you may of course contact him directly), but an extract was shown e.g. in a conference paper  (pp. 676-677).
- Schregle, Renken, and Wittkopf (2020). Spatio-temporal visualisation of reflections from building integrated photovoltaics. Buildings, 8(8). http://dx.doi.org/10.3390/buildings8080101
- Noback (2020). Tageslichtsimulation verlorener Gebäudezustände am Beispiel der Hagia Sophia Justinians. Ph. D. dissertation, Technische Universität Darmstadt, Germany.
- Grobe, Noback, and Inanici (2020). Challenges in the simulation of the daylight distribution in late antique Hagia Sophia. In: International Hagia Sophia Symposium, Istanbul, Turkey. http://hdl.handle.net/11352/3239