基于粒子滤波的自组织模糊神经网络算法研究.pdfVIP

学兔兔 第32卷 第3期 仪 器 仪 表 学 报 V0I．32 No．3 2011年3月 Chinese Journal of Scientific Instrument Mar．2011 基于粒子滤波的自组织模糊神经网络 算法研究 程洪炳，黄国荣，倪世宏，刘华伟 (空军丁程大学工程学院 西安 710038) 摘 要：为了得到结构更加紧凑、泛化性能更强的自组织模糊神经网络，提出了基于粒子滤波(particle filter，PF)的自组织模糊 神经网络训练算法。其能够对模糊规则进行自动生成和增删。文中给出了模糊规则生成准则，应用误差率下降方法作为模糊 规则增删策略，删除作用不大的规则。建立了以隶属函数宽度参数为状态，以理想输出为量测的动力学模型，利用PF对参数进 行了学习。最后，对两个实例进行了仿真，从仿真结果可以看出，与D．FNN、SOFNN、EKF．SOFNN等算法相比，其在结构紧凑性 以及泛化性能上都得到了提高，从而证明了PF。SOFNN的有效性。 关键词：粒子滤波；自组织模糊神经网络；误差率下降；模糊规则；隶属函数 中图分类号：TN391．4l 文献标识码：A 国家标准学科分类代码：510．40 Design of self-organizing fuzzy neural network based on particle filter Cheng Hongbing，Huang Guorong，Ni Shihong，Liu Huawei (Engineering College，Air Force Engineering University，Xi’an 710038，China) Abstract：In order to obtain self—organizing fuzzy neural network(SOFNN)with more tightened structure and better generalization，a SOFNN training algorithm based on particle filter is designed and developed．Firstly，the generation criterion of fuzzy rules is given．The fuzzy rules can be generated，added and deleted automatically．Then，the meth— od of error ratio reduction is used as fuzzy rule pruning strategy：the rule that plays an unimportant role in the system is deleted．A dynamic model that takes the width of membership function as the state and takes the ideal output as the measure is established．Particle filter is used to learn the parameters．Final