Industrial control federal intrusion detection backdoor defense method and system based on clustering combination optimization

The invention discloses an industrial control federal intrusion detection backdoor defense method and system based on clustering combinatorial optimization, and the method comprises the steps: generating a malicious update gradient through employing a plurality of conventional backdoor attack methods, and enabling the malicious update gradient and a normal update gradient to serve as a training set and a test set; a combination mode of multiple clustering methods is used as an optimization variable for coding, a training set is used for training the combination mode, the true positive rate and the true negative rate of malicious gradient recognition of the combination mode on a test set are evaluated, the true positive rate and the true negative rate serve as optimization objective functions, and then a clustering combination mode with high recognition accuracy is obtained through multi-objective population evolution operation. And the method is used for a backdoor attack defense strategy in an industrial con