Research And Implementation Of Industrial Internet Of Things Device Identification Technology Based On WEB Information And Machine Learning

At present,hundreds of millions of industrial IoT(IIoT)devices are exposed on the public Internet.However,most of the equipment manufacturers have not prioritized product security due to cost savings,coupled with factors such as low user safety awareness and poor equipment management.So hackers can easily control a large number of devices,which triggers large-scale attacks such as the "Mirai" virus sweeping the world and causes irreparable losses.Therefore,it is imperative to investigate the hidden safety hazards of IoT devices and prevent IIoT security threats.Because IIoT devices of the same type,the same series,or from the same brand often have the same or similar security vulnerabilities,it becomes very important to identify IIoT devices in the cyberspace,including their types,brands,and models.Because the distribution of overall IoT devices in the network space has a long-tail effect,and the methods of identifying industrial routers,industrial firewalls and other devices are the same as the methods of identifying common IoT devices.Therefore,this paper proposes a scheme to construct different mathematical models for common IoT devices and emerging IIoT devices for fine-grained device identification.This paper starts from the web management page of IoT devices,and focuses on the characteristics of common types of IoT devices at first,such as routers and cameras.Because these devices are relatively concentrated,with a large number of each type of device,and the information about the device model,vendor,etc.in the body of the device's Web homepage is obscure,a scheme for supervised learning based on multi-feature fusion for device classification is proposed.This solution integrates multi-angle features such as Banner,text content,hyperlinks,and page structure of Web management pages,which deeply describes common types of IoT device types.This method is more universal than previous research work,and make up for the defect that the single feature or the integrated text feature sometimes cannot identify some devices.This article uses a combination of Random Forest and improved CHI feature extraction algorithm to describe the common types of IoT devices.The accuracy of the resulting model for all types of IoT devices is above 98.3%.The automatic labeling of vendors and models of common IoT devices is based on the establishment of a Trie-based device information database.In recent years,the emerging types of IIoT devices are numerous and complex,and are relatively fragmented.Moreover,the number of accessible industrial IoT devices in the real network environment is small,but the information about device vendors and models in the Web management page is obvious.Therefore,it is not suitable to use classification methods for IIoT device identification.This paper proposes a scheme for mining device fine-grained identification rules based on the seven key fields of the Web management page and Apriori algorithm,which expands the device identification range.Finally,the similarity is used to match the rules of the device to generate(type,brand,model)label.After experiments,the accuracy of the method reached 99.9%,and the recall rate reached 97.3%.