基于高阶谱约简的变结构模糊神经网络

针对先验知识不完备和不确定的情况下海量数据造成的冗余和互斥，模糊神经网络结构变得复杂化并不能很快逼近和分类输出对象的情况，提出了一种基于高阶谱完成规则约简的变结构模糊神经网络的模型。相同结论属性的模糊规则的条件属性值可以被认为是由若干个谐波成分组成的平稳信号，并且此信号可以采用高阶谱分析来估计其谐波成分，规则的最小约简集与谐波对应。在完成了谐波估计后，神经网络结构和连接权值发生改变，神经网络的性能也得到优化。最后给出了此模型在航迹融合中应用的一个例子，得到了较好的结果。

Structure Changed FNN Based on High Order Spectral Reduction

With the structure of FNN turning more complex,the good approach or classification is hard to get because of redundancies and uncertainties caused by both large volume uncertain data and lack of prior knowledge.Based on high order spectral reduction for rulers,a new structure changed Fuzzy Neural Network(scFNN) is proposed.The value of attribution in fuzzy rulers having same result can be considered as a stationary signal set composed by several harmonics,and the signal can be analyzed using high order statistics to estimate the harmonics.The minimum reduction set of rulers is corresponded to the harmonics.Then the structure and the joint weight of FNN can be changed after harmonics estimation.The performance of scFNN is also optimized.Finally,the model is used in track-to-track fusion and a good result is obtained.