Intrusion Detection Model Optimization Scheme For Gradient Boosting Trees In Edge Computing

The Industrial Internet of Things uses edge computing,artificial intelligence,cloud computing and other technologies to combine with future advanced manufacturing technologies to build a new intelligent production system.By sinking ICT infrastructure,the interconnection of edge equipment,protocols and data is realized,providing computing power for industrial manufacturing to process data at the edge.The edge computing center is close to the user side,which can process the perception data nearby,without sending it to the cloud.However,the lack of perfect data encryption and data protection measures makes it easy to be invaded by attackers in the process of processing localized sensing data.Intrusion detection technologies based on machine learning provide strong security for edge computing center,in which the most widely used is Light Gradient Boosting Decision Tree(i.e.,LGBDT).But still this model faces with problems such as imbalanced data,high dimensional data characteristics,and low efficiency of parameter optimization.To solve these problems,this thesis proposed an optimization scheme for LGBDT to improve its detection precision and training efficiency.First,to achieve the data class balance in data set,we proposed a gradient based sampling borderline synthetic minority(i.e.,GSBSM),which first updates the data according to the sample gradient value and deletes the small gradient sample,and then to expand the non-noise data with less sample size,namely,small sample,to ensure equilibrium distribution of data.Second,to reduce the number of data feature,we proposed exclusive features binding-hierarchy cross validation algorithm(i.e.,EFB-HCV).According to the principle of mutual exclusion,the new algorithm combines features,reduces the number of features,and strengthens the relationship between features and goals.It also designed hierarchy system to ensure equal proportionment of data category(attack category)in training set and testing set at cross validation stage.Next,in order to improve the efficiency of parameter optimization in model training process,we proposed gaussian distribution based bayesian optimization(i.e.,GDBO)to improve retrieval efficiency of optimum parameters.Finally,the detailed experimental results showed that our new scheme achieves the data balance in the data set,reduces the data characteristic dimension and retains the valid information,and improves the efficiency of parameter optimization and the model performance under the optimal parameters.Moreover,the new scheme defended against intrusion more effectively.