I believer the load should be just:
int_A q_z * e_z * n * dA where q_z is load intensity, e_z the unit vector (0,0,1), n the normal to the shell and dA the area. In isoparametric coordinates n*dA is just dx/dxi x dx/deta where xi and eta are the parent coordinates and x is the coordinates in cartesian form x = x_A N_A(xi,eta) the N_A are bilinear. The loading can be "dead" where x is just the reference coordinates at time 0 or it can be current where x = X + u -- in the later case there will be a tangent also.
Look in any FEA book where normal loading is computed (e.g., see Zienkiewicz, Taylor and Zhu).