how to choose compensations for loop stability?

As you know, if you should design a good controll sysytem, you must keep the sysytem stable at first and then you will use frequency compentstion technology for better performance. you choose some compensated parameters,just the value of C, R and so on, later you use simulation and get the loop gain-phase figure, then, you decide which case is the best one. Do you realy do so? But the fact is always different with the simulation, how can you be sure that your design have a good real performance. I do not know how to design a good loop.......

I always use this way. If your simulation is so different from reality you must think over what is wrong with your models. If you find spice model somewhere in Internet and use this model of couse your simulation will be different from measurement. But if you use the models from your foundry the simulation should be very close to reality.
I remember once I had made design change without simulation (it was change only one of the block of whole device and I simulated this block behavior only because I had no time) and after all I get in real device parasitic oscillation of another block. After that I simulated this another block and get that for both AC and transient analysis it should oscillate.
I also don't know universal remedy for frequency compensation. It depends on a lot of factors. Some methods cannot be used for low supply voltage, another increase current consuption. If you design a LDO with limitation for both of them frequency compensation become a very difficult task, because there are others destabilizing factors: heavy capacitance loading and current loading that can be changed form 0 to max value.

You can have enough margin for process variation. If you go through with all process corner, and a monte carlo, it will be very sure.