Research On Intrusion Detection Strategy For IIOT Based On Self-Encoding Multi-layer ELM

With the rapid development of computer technology and information technology,the Industrial Internet of Things(IIoT)based on these technologies has also developed rapidly.But the development also brings security issues.Once the IIoT suffer a cyber-attack,it may lead to the tampering of production equipment parameters,resulting in major production accidents,or the leakage of enterprise production data,resulting in significant losses to the interests of the enterprise.Therefore,industrial IoT security issues need to be paid great attention.For the increasing number of industrial IoT attacks,researchers have designed an Intrusion Detection System(IDS)to detect the security issues of the industrial IoT in real time.The information transmission between the terminals of the IIoT needs to be realized through an open network,which makes it easy for attackers to control the entire IIoT system by attacking a weak link in the network.So,the IIoT intrusion detection system must have good real-time,accuracy and computing capabilities.The multi-layer ELM autoencoder has high training efficiency and good model performance,which enables real-time intrusion detection for the IIoT.Therefore,this thesis studies the security strategy of the industrial IoT intrusion detection system based on the Multi Layer ELM Auto Encoder(ML-ELM-AE).(1)An intrusion detection strategy of ML-ELM-AE based on L₁penalty-item embedded feature selection algorithm is proposed.The data volume of the IIoT is large,the feature dimension is high,and there are many redundant features.Therefore,an embedded feature selection algorithm based onL₁penalty terms--L1-EM is proposed.The feature selection algorithm can calculate the weight of each feature,and then select the feature with a larger weight according to the weight,thereby reducing the redundant features in the original data set and reducing the feature dimension.Then the L1-EM algorithm is combined with the Multi Layer ELM Auto Encoder,and the L1-EM-MLELM intrusion detection algorithm is proposed,thereby improving the performance of Multi Layer ELM Auto Encoder training model.(2)An intrusion detection algorithm based on particle swarm optimization algorithm is proposed.The input weights of the original ML-ELM-AE and the biases of the hidden layer neurons are randomly initialized,and the performance of the training model is closely related to the values of these two parameters,which leads to the randomness of the performance of the ELM.Therefore,in this thesis,particle swarm optimization algorithm is used to find the optimal input weights and hidden layer neuron biases.The improved algorithm improves the performance of the training model.