An Intrusion Detection Model Of Internet Of Things Based On Improved CNN And RNN

Recently,with the growth of the scale of the Internet of Things and its popularization in many aspects of life,the character of the Internet of Things has turned increasingly important,but this also means that the Internet of Things devices are more likely to be considered as the target of network attacks.Intrusion detection systems detect network traffic in actural time.When an attack is detected,the intrusion detection system will issue an alert in a timely manner,thereby achieving the goal of ensuring network information security.Most Internet of Things devices,compared to traditional network devices,may need to be deployed in environments lacking human maintenance.Internet of Things devices are not only vulnerable to external damage,but also vulnerable to network attacks due to their limited computing power and storage capabilities.Therefore,in addition to focusing on improving model performance,it is also necessary to consider whether the built model is suitable for the Internet of Things environment.This paper proposes an improved intrusion detection model of the Internet of Things,based on the extant intrusion detection models which are based on the Convolution and Recurrent Neural Network.The proposed model is made of a feature extraction model based on the improved Convolution Neural Network and a classification model based on the improved Recurrent Neural Network.The main research are as follows:(1)Aiming at the characteristics of small memory and weak computing power of the Internet of Things devices and the problem of information loss in traditional pooling layers,this paper proposes a feature extraction model based on an improved Convolutional Neural Networks.The feature extraction model includes two parts of improvements.One is that the convolutional layer uses the small convolutional core,which greatly reduces the parameter amount of the convolutional layer.In order to ensure the same perception field as the large convolutional core,multi-layer convolutional layers are stacked.The second is that the pooling layer in the model uses the Soft Pool pooling layer to replace the original pooling layer,achieving feature dimensionality reduction.(2)Aiming at the problem of gradient disappearance when processing the longsequence data by the Recurrent Neural Network,this paper proposes a classification model based on an improved Recurrent Neural Network.In the classification model,the recurrent layer uses the Simple Recurrent Unit,which has relatively few parameters and a simple calculation process,and also implements parallel computation between different time steps.And the Simple Recurrent Unit also is combined simple with multi-head structure for analyzing data from multiple perspectives.(3)The performance of the proposed model is verified through the simulation experiments.In the simulation test,the Internet of Things data sets Bo T-Io T and TONIo T are used as the benchmark data sets,and the evaluation indicators used include accuracy,detection rate,F1-score,precision,and false alarm rate.The simulation test includes the ablation study related to the proposed model,and compares the performance of the proposed model with the performance of existing models.The experimental results show that the proposed model can control the parameters within a small range,while the improvements made based on convolution and recurrent neural networks can help improve the performance of the intrusion detection model,which has practical significance.