Research On Intrusion Detection Of Industrial Internet Of Things Based On Support Vector Machine

The Industrial Internet of Things is a world-wide transformation of the potential of the digital industry,which has brought earth-shaking changes to the survival and life of mankind.In the face of the continuous update of industrial physical network equipment and technology,how can data be protected from threats has become a hot issue,so intrusion attacks have become one of the main problems of Industrial Internet of Things network security.From the bottom to the top,Industrial Internet of things usually has three layers:physical layer,communication layer and application layer.There are not only a lot of data transmission,sharing,operation and analysis between each layer,but also a lot of data connection and coupling between layers.In the Industrial Internet of things,the data communication between massive devices all the time also provides opportunities for many malicious data.Take the communication layer of the Industrial Internet of Things as an example,it is the key layer that connects the Industrial Internet of Things platform and sensors or controllers and it is also the most standardized,strongest and most mature part of the entire industrial physical network architecture.As a result of the development of industrialization,the security issues of the communication layer of the Industrial Internet of Things will receive more attention and face more challenges.Aiming at the huge security risks faced by the Industrial Internet of Things during its operation,this article carefully understands the current intrusion problems faced by the Industrial Internet of Things on the basis of the Industrial Internet of Things.Through the application of relevant algorithms of machine learning,the Industrial Internet of Things based on support vector machines is developed.In the research of networked intrusion detection,two intrusion detection methods and corresponding workflows have been researched.Experiments have proved that these two methods can well detect malicious intrusion in the Industrial Internet of Things.This article specifically conducts research from the following two aspects:1.In offline intrusion detection,all types of data can only be analyzed after all the information is forwarded and transmitted.Therefore,a single vector machine cannot satisfy the detection of many types of abnormal data.In response to this type of problem,we consider using different support vector machines to identify and detect different types of data according to the frequency of malicious data,and propose a dualdimensional reduction intrusion detection method based on support vector machines.Specifically,we first perform data preprocessing on the acquired data,including numerical type conversion and data normalization,and then use pearson correlation coefficient and Light GBM to double-extract the features of the data,so as to transform high-dimensional data into low dimensional data without losing the detection accuracy of the model.Secondly,we input the data into the model based on support vector machines to realize the detection and recognition of malicious data.Finally,experiments are carried out with NSL-KDD as a data set,and the results are compared with many new intrusion detection methods,which proves the superiority of our method in intrusion detection.2.Online intrusion detection can quickly respond according to the actual trend of data,but it cannot accurately count the frequency of a certain type of data,so the intrusion detection method involving multiple support vector machines is no longer applicable.Based on this,we propose an intrusion detection method involving a single support vector machine,PSO-Light GBM.Based on the pearson correlation coefficient and Light GBM,we further introduce the binary particle swarm optimization(BPSO)algorithm to make the extracted features more accurately describe the entire data set.In the model building stage,we only use one class support vector machine(OCSVM)for training,avoiding the inconvenience of using multiple support vector machines.In the experimental verification of UNSWNB15 as the data set,the experimental results show that the intrusion detection method proposed by us has high accuracy in "Normal" and various malicious data,especially in the detection of small sample data such as "Backdoor","Shellcode" and "Worms".At the same time,the false alarm rate and the training and detection time of the model also show excellent performance.