基于平淡粒子滤波的自组织模糊神经网络算法

摘要

In this paper, a Self-Organizing Fuzzy Neural Network (SOFNN) based on Unscented Particle Filter (UPF) was designed and developed. The UPF was used to estimate the parameters of the SOFNN and better result was gotten. The generating criterion of fuzzy rules based on the pruning strategy of the error reduction ratio was introduced. The width of membership function was established as the state and the ideal output as the measurement. The UPF was used to leam parameters. The two typical simulations, nonlinear function approximation and system identification, were done to validate the UPF-SOFNN. It can be seen from the results of simulation that the UPF-SOFNN has a more compact structure and better generalization than the other algorithms.%提出一种基于平淡粒子滤波(UPF)的自组织模糊神经网络(SOFNN)训练算法——UPF-SOFNN.分析了基于误差下降率的模糊规则增删策略和神经元增加和删除准则,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,并利用UPF对神经元参数进行学习.分别以非线性系统函数逼近和系统辨识为例,对UPF-SOFNN算法进行了仿真验证,表明UPF-SOFNN算法具有更紧凑的结构和较强的泛化性能.