Research On Industrial Control Intrusion Detection Method Based On Ensemble Learning And Kernel Extreme Learning Machine

With the introduction and implementation of the national industry 4.0 strategy,although the industrial control(industrial control)system can bring convenience to it through the integration of the traditional Internet,it also brings more security risks to the industrial control system.Intrusion detection method is to identify the attack behavior in the system and give the corresponding response strategy in order to eliminate the hidden danger of system security.It is widely used in the field of industrial control security.Machine learning is a technology for developing computer algorithms that can simulate human intelligence.Kernel Etreme Learning Machine(KELM)is one of the classic methods of machine learning.This model has the advantages of fewer parameters,fewer iterations and higher operating efficiency,and is widely used in the field of intrusion detection.However,the existing KELM models still have problems such as the difficulty of finding the best parameters,and the classification accuracy of a single KELM model is not high.In view of the shortcomings of the KELM model,this paper mainly proposes the optimal model TCDA-RF-SEKELM in three stages.The specific research contents and innovations are as follows:1.A Tabu Search Chaotic Differential Evolution Artificial Bee Colony(TCDA)algorithm is proposed in this paper.The algorithm first uses adaptive mutation factor and elite evolution strategy to improve Differential Evolution(DE)algorithm,and then uses tabu table to improve Artificial Bee Colony(ABC)algorithm.Then,based on the leapfrog idea,the above two improved algorithms are merged,and chaotic initialization mechanism is introduced to form the final TCDA algorithm.Finally,this paper uses the standard test function to experiment the TCDA algorithm,and the simulation results show that,compared with other algorithms,the TCDA algorithm achieves the best computational accuracy and operating efficiency.2.In view of the shortcoming that the best parameters in KELM are not easy to find,this paper uses the TCDA algorithm to optimize the KELM model and obtain the TCDA-KELM model with the best classification performance.At the same time,for the single classifier model detection accuracy bottleneck problem,first introduce the Rotation Forest(RF)integrated learning model is used to obtain a subset of the data with greater differences,and then the better-performing TCDA-KELM model is used as the base classifier to build the TCDA-RF-EKELM integrated learning model.The results show that the detection accuracy of TCDA-RF-EKELM model can reach96.24%,which is 2.28% higher than that of single TCDA-KELM classifier in the University of Mississippi's power industrial intrusion detection data set.3.Aiming at the defects of the TCDA-RF-EKELM ensemble learning model,such as low detection rate and large storage space occupied,this paper carried out selective integration of the TCDA-RF-EKELM model to eliminate the base classifier with poor classification effect and only retain the base classifier with good classification effect.Finally,the TCDA-RF-SEKELM integrated model with better overall performance is obtained.The results show that TCDA-RF-Sekelm reduces the running time by 43.5%compared with TCDA-RF-EKELM,and the model has the highest detection accuracy of 96.86%,which is a better industrial control intrusion detection model.