As far as I know the lumped port attempts to simulate an excitation probe on your circuit.
I have used a lumped port on cpw structure to simulate my sample.
I have used for the port a width equal to the width of the cpw signal trace.
I left the same gaps of the cpw line and I connected the opposite side to some extended grounds.
I also simulate the same structure with a waveport instead of a lumped port.
After fabricating my sample and tested on a probe station, I realized that the results coming from the simulation with a lumped port was more close to the real measurements than the simulations with waveport.
Moreover, changing the lumped port size in the simulation was actually giving a slightly different results.
By the way, The result that was close to the real measurements was the one I just suggested above.
So, just add to the input of the cpw signal trace a square whose side is equal to the cpw trace width and thatn after close behind this square the two grounds.
Don't forget the integration line which goes from the ground to the trace.
A bit of a brain dump but hope it helps.