免疫连续记忆故障诊断方法  

Immune Continuous Memory Fault Diagnosis Method

作　　者：张宏利[1] 兰超 刘树林[1] 肖海华 蒋伦常 孙欣 ZHANG Hongli;LAN Chao;LIU Shulin;XIAO Haihua;JIANG Lunchang;SUN Xin(School of Mechatronic Engineering and Automation,Shanghai University,Shanghai 200444)

机构地区：[1]上海大学机电工程与自动化学院,上海200444

出　　处：《机械工程学报》2023年第24期34-45,共12页Journal of Mechanical Engineering

基　　金：国家自然科学基金(52075310,61603238);上海市科技创新行动计划(22142200102)资助项目。

摘　　要：故障样本少且难以获得是制约智能诊断系统发展的关键问题,基于生物免疫系统的克隆、变异及连续学习等众多智能机理,提出了一种具备新增故障样本在线连续学习能力的免疫连续记忆算法(Immune continuous memory,ICM)。算法采用高斯函数计算空间任一点故障样本(抗原)出现可能性的大小称为细胞因子浓度,并根据细胞因子浓度大小提出尺度可变B细胞(Scale variable B cells,SVB)概念,模拟生物免疫克隆变异机理针对任一抗原形成数量较多但识别能力更强的成熟B细胞,进而提出一种记忆细胞优选策略,对成熟B细胞数目进行优化,形成数量较少且能够识别所有训练抗原的记忆B细胞群用于数据分类。构建记忆B细胞群条件更新机制,对新增抗原进行持续学习,形成识别能力更强的记忆B细胞集,实现ICM的连续学习功能。经标准数据集仿真表明,所提出的ICM算法在同等条件下与其它算法相比具有较好的分类性能。经往复压缩机气阀诊断试验证明,ICM算法能够通过不断更新记忆细胞提高其诊断性能,是一种有效的连续学习故障算法。The lack of fault samples and the difficulty of obtaining them are the key problems that restrict the development of intelligent diagnosis systems.Based on the numerous intelligent mechanisms of biological immune system,such as cloning,mutation and continuous learning,an immune continuous memory algorithm(ICM)with the ability of online continuous learning of new fault samples is proposed.This algorithm uses Gaussian function to calculate the probability of the occurrence of faulty samples(antigens)at any point in space,which is called the cytokine concentration.The concept of scale-variable B cells(SVB)is proposed according to the concentration of cytokines and the mechanism of biological immune clonal variation is simulated to form mature B cells with more number but stronger recognition ability for any antigen.Then a memory cell optimization strategy is proposed to optimize the number of mature B cells to form a small population of memory B cells capable of recognizing all training antigens for data classification.The condition update mechanism of memory B cell population is constructed to continuously learn new antigens and the memory B cell set with stronger recognition ability is formed to realize the continuous learning function of ICM.Simulation results on standard datasets show that the proposed ICM algorithm has better classification performance than other algorithms under the same conditions.The experimental results show that the ICM algorithm can improve the diagnostic performance by constantly updating memory cells and it is an effective continuous learning fault algorithm.

关 键 词：故障诊断 连续学习 生物免疫系统 优选记忆细胞 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]