Lol. Lmao.This is exactly the kind of problem I want to give to a deep learning + reinforcement learning system and just have it play it out over and over and over again. It should be able to devise optimal strategies to employ IADS assets and identify the most vulnerable approach routes. I hope and expect that the PLA is doing this.
You should look up what a Generative Adversarial Model (also called GAN) is. We uh, well... we may or may not base course of action analysis on deep-learning-ifying a GAN model by giving it force postures, objectives, etc. etc. and having it run force employment models hundreds of thousands (if not millions LOL) of times, and using it to determine what the optimal employment schema for a given operational environment are. I also may or may not have used that approach when working on JAAWS (joint ai augmented weaponeering system), which takes force structures, ETFs (electronic target folders), munition distributions and current taskings, and can work out the most efficient munition allocation and conemp.
EVERYBODY is doing this kind of thing. I'd almost bet money that the PLA has this, but significantly better (afaik JAAWS only ended up getting used in Cruise Missile Support Activity - Atlantic as some auxiliary thing because it takes a while to run and can be finicky) in almost every way. Data is the future of combat (and everything else, really); and the side with access to the best data and the best tools to work with that data is the side that will perform best in war. After all, the PLA doesn't define it's 3 major theaters as "Air, Sea, and Information" for no reason.