Research On Data Link Anomaly Detection Model Based On Machine Learning

With the large-scale application of the industrial Internet,data link attacks on the industrial Internet pose a huge threat to industrial production.In today’s complex and diverse network attacks,using only proxy,gateway,firewall,encrypted tunneling,and other technologies can no longer prevent these malicious network attacks.Therefore,the active identification of data link anomalies has become a research hotspot in data security.The existing research results mostly detect data link anomalies in sample balanced situations,but the effect of data link anomaly detection in sample unbalanced situations is very unsatisfactory.Therefore,this article focuses on the following work to address this defect in traditional anomaly detection:Firstly,this article proposes an improved sparrow search algorithm and uses its optimized deep belief network to extract various abnormal features in the link,enhancing its classification ability during sample balancing.This article introduces three-point optimization based on the problems in the sparrow search algorithm,which solves the defects of the algorithm being prone to falling into local optima and unable to find the optimal solution to the problem.Through experiments in a preprocessed CICIDS201 7 dataset,it is found that the improved sparrow search algorithm optimized deep belief network outperforms other similar methods in terms of accuracy and precision when performing classification after extracting the characteristics of various anomalies in the link.Secondly,through comparative experiments,it is found that fuzzy extreme learning machine has better classification performance when faced with minority classes,outliers,and noise samples.Therefore,this article will use the improved dung beetle optimizer to optimize the fuzzy extreme learning machine.According to the defects in dung beetle optimizer,three-point optimization is introduced,combining weighted voting with a fuzzy extreme learning machine optimized by an improved dung beetle optimization algorithm,WV-IDBO-FELM is proposed.By adding corresponding weights to each classifier and making full use of each layer of information in the deep belief network,the disadvantage of only having a unique classifier can be overcome.Through experiments in non-sample balanced NSL-KDD data sets,it is found that WV-IDBO-FELM has more advantages in accuracy and false alarm rate than IDBO-FELM when distinguishing various anomalies in the link.Finally,this paper combines the deep belief network optimized by the improved sparrow search algorithm with the fuzzy extreme learning machine optimized by the improved dung beetle optimization algorithm incorporating weighted voting to propose a link data anomaly detection method under non-sample equilibrium,namely DBNWV-IDBO-FELM.It utilizes the advantages of IDBO and improved SSA in terms of optimization time,accuracy,and balancing the relationship between local and global exploration.It also takes advantage of the deep belief network in feature extraction.At the same time,adding weighted voting compensates for the disadvantage of only having a unique classifier.After verification,the DBN-WV-IDBO-FELM proposed in this article performs well in accuracy,precision,F1 value,and false alarm rate.