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。 Aiming at the redundancy and mutual exclusion caused by massive data under the circumstance of incomplete and uncertain prior knowledge, the structure of fuzzy neural network becomes complicated and the object of output can not be approximated and classified quickly. Variable structure fuzzy neural network model to complete rule reduction. Conditional attribute values ​​of fuzzy rules of the same conclusion attribute can be considered as a stationary signal composed of several harmonic components, and this signal can be estimated using high-order spectral analysis of its harmonic components, the rule minimum reduction set and harmonics correspond. After the completion of the harmonic estimation, the structure of the neural network and the connection weights are changed, and the performance of the neural network is also optimized. Finally, an example of the application of this model in track fusion is given, and good results are obtained.