The incidence of postoperative delirium in elderly patients is high, which can lead to long-term postoperative neurocognitive disorders. Its high risk factors are not yet clear. At present, there is a lack of early diagnosis and alarm technology for perioperative neurocognitive disorders, which can not achieve early intervention and effective treatment. By artificial intelligence and autonomously evolutionary neural network algorithm, relying on multi-source clinical big data, we explored the use of Bayesian network to optimize the anesthesia decision-making system in enhanced recovery after surgery, and established risk prediction model for perioperative critical events. It is expected that this method will also help to establish a risk prediction model for postoperative delirium and long-term postoperative neurocognitive disorders. This project plans to collect the perioperative sensitive parameters of anesthesia machine, multi-parameter monitor, EEG monitor,fMRI and HIS system, to explore the evolution process of data characteristics by feature fusion.We also plan to quickly screen key perioperative risk characteristics of postoperative delirium from massive clinical data through feature selection, to explore the high risk factors of long-term postoperative neurocognitive disorders developing from postoperative delirium. Finally, with multi-center intelligent analysis，the risk prediction model of postoperative delirium and long-term postoperative neurocognitive disorders will be constructed.