A Study Of Dendritic Cell Algorithm Oriented Anomail Detection

Dendritic Cell Algorithm(DCA)is based on biological function of dendritic cells.It has the advantages of small computational scale,strong recognition ability and no need for a large number of training samples,so it has been successfully applied to anomaly detection,image classification and fault diagnosis.However,due to its short time,there is still much room for research in optimization,improvement and application.Based on this,this paper focuses on the optimization and improvement of DCA,and applies it to anomaly detection.The main tasks are as follows:1.An improved dendritic cell algorithm for unordered data setsAiming at the fact that DCA is sensitive to the order of input data,which results in poor detection performance on unordered data sets,an improved DCA for unordered data sets is proposed.Firstly,the reason why DCA is sensitive to the order of input data is analyzed.Secondly,the antigens and output signals are fused as input data,and each DC is required to collect only one class of data with the same number,thus avoiding the mutual interference between the data.Finally,the antigen was evaluated according to the value of cell environment.The experimental results show that the detection performance of the algorithm is not affected by the data sequence,and the detection performance on the unordered dataset is better than the standard DCA and other methods,its accuracy,false positive rate and false alarm rate are respectively 91.14%,5.29%and 3.57%.2.An improved dendritic cell algorithm based on principal component analysisIn order to realize automatic dimensionality reduction and improve the detection performance of the algorithm on unordered data sets,an improved DCA based on principal component analysis is proposed by introducing the principal component analysis method and establishing the corresponding relationship between the selected principal components and the input signals.Firstly,the z-score method is used to standardize the dataset,and the standardized dataset is reduced by the principal component analysis method.Secondly,based on the contribution rate of the principal component and the influence weight of the input signals to the output signals,the corresponding relationship between the selected principal components and the input signals is established,and each selected principal component is assigned to the specific signal category.Finally,the influence of principal component number,weight matrix and signal transformation formula on the performance of the algorithm is analyzed by controlling parameters.The experimental results show that the algorithm achieves automatic dimensionality reduction and the detection performance is also improved,its accuracy,false positive rate and false alarm rate are respectively 96.71%,0.86%and 2.43%.