I would be interested to find out how this problem gets resolved. In my work developing the IWEC2 weather files for ASHRAE, I encountered problems applying the Perez model to get the direct/diffuse split, and ended up switching to a simpler Gompertz Function model.