Study On Artificial Immune System Based On Multi-granularity Immunoregulatory Mechanism And Its Application In Fault Detection

With the increasing scale and complexity of modern industrial,the probability of equipment failures in production process has greatly increased.In order to reduce equipment loss and accidents,implementing effective fault detection on equipment and systems has become an important problem in modern industry.The Artificial immune system based on biological immune mechanism belongs to the data-driven artificial intelligence method and has been widely used in the field of fault detection.It has intelligent features such as self-learning,self-adaptation,and pattern recognition,as well as high execution efficiency and the ability to identify unknown threats.However,the biological immune mechanism referenced by artificial immune system is single and imperfect,which does not fully reflect the adaptive characteristics of biological immune system.As a result,the failure detection effect is not good when faced with a complex and changeable system environment,so the adaptability of artificial immune system needs to be further enhanced.The immune regulation mechanism is the key to the adaptive ability of immune system,which exists in all aspects of immune response process.This paper introduces the idea and mechanism of multi-granularity immunoregulation of immune system to improve the adaptability of artificial immune system from different levels of regulation,so as to improve the fault detection ability of artificial immune system in industrial environment.Therefore,the main research work of this paper includes:In order to solve the problem of dynamicity and hole-coverage of negative selection algorithm(NSA),this paper proposes a NSA based on miRNA regulation mechanism——miNSA.At the gene level,the miRNA regulatory mechanism of T cell development is introduced.The original NS model is improved from three aspects of proliferation regulation,negative regulation,and TCR affinity regulation,which are mapped to the adjustment of number and radius of mature detectors.Through real-time judgment of false positives and false negatives,and analysis of spatial coverage of detection data between detectors and self-set,the dynamic adjustment of radius and number of detectors is achieved.Therefore,the adaptability of NSA for fault detection is improved.By testing on the underwater junction box of submarine observation network,the detection rate and false positive rate of miNSA on abrupt fault are better than NSA algorithm.In view of the fact that dendritic cell algorithm(DCA)relies heavily on artificial experience to set parameters,this paper proposes a DCA based on cytokines(CKs)feedbackregulation——ckDCA.At the molecular level,the feedbackregulation mechanism of DC antigen presentation by CKs is introduced.The original DC model is improved from two aspects of maturation differentiation regulation and migration ability regulation,which are mapped to the adjustment of signal conversion weight and migration threshold,meanwhile the CK evaluation index is proposed as the basis for the feedbackadjustment.In addition,in terms of signal definition,an adaptive extraction method of antigen signals based on sorting of coefficient of variation is proposed.By testing on the DAMADICS,although the detection rate and false positive rate of ckDCA on abrupt fault and incipient fault are slightly better than DCA,ckDCA detects faster on incipient fault.Aiming at the lackof generality of danger definition and integrity of danger assessment in complex system environment,this paper proposes an immune homeostasis method based on T cell response regulation——IHDC-FD.In the danger definition,based on the immune homeostasis mechanism,a method of danger perception based on change is proposed.The changes caused by system imbalance are considered to be dangerous external manifestations of system.In order to achieve the adaptive extraction of antigen signals,the first-order derivative in numerical differentiation is used to describe the change,and the second-order derivative is used to determine whether the change represented by the first-order derivative is normal or abnormal.In the danger assessment,based on the T-cell response regulation mechanism at the cellular level,a system balance assessment method based on the Homeostasis Index(HI)is proposed.The concentration of Th and Treg that regulate the immune homeostasis is used to calculate HI to reflect the degree of system imbalance for the overall health evaluation of system.Through testing on the TE,IHDC-FD has better fault detection performance than DCA and PCA,especially on incipient fault it features high detection rate and low detection delay.Based on previous work,by introducing the hierarchical design ideas of large-scale system monitoring,this paper constructs a fault detection immune hierarchy based on the multi-granularity immunoregulation mechanism for large-scale system.The hierarchy consists of immune entity and immune decision center.The immune entity uses two methods: DC antigen presentation and T lymphocytes to implement fault detection of subsystem.The immune decision center is responsible for collecting and analyzing the reporting information of each immune entity,feedbackand adjustment of the operating parameters of immune entity to achieve the self-adaptation of antigen detection.Meanwhile,it uses the shared dangerous antigens for training to generate T cell detectors,so that each immune entity can quickly detect similar dangerous antigens to achieve the purpose of herd immunity.Through the above research,the multi-granularity immunoregulation methods proposed in this paper further enhance the adaptability of artificial immune system and have been experimentally verified to have better fault detection performance in industrial environment.