摘要
在D-FNN算法基础上,提出了基于椭圆基函数(EBF)的广义动态模糊神经网络。该算法提出模糊ε-完备性作为高斯函数宽度的确定准则,避免初始化选择的随机性;同时,该算法不仅能对模糊规则而且能对输入变量的重要性作出评价,从而使得每个输入变量和模糊规则都可以根据误差减少率(ERR)来修正。其应用不仅可以用来建模,还可以用来抽取有意义的模糊规则以获取知识。通过与D-FNN以及其他方法的比较,可以看到GD-FNN在学习效率和性能方面具有突出的优势。最后针对实际案例进行了仿真分析,验证了该算法的有效性和高效性。
According to the D-FNN algorithm, the general dynamic fuzzy neural network was proposed based on the ellipse primary function （EBF）. In the algorithm, the fuzzy Epsilon-completeness was taken as the definite criterion of width for gaussian function, by which the random choice for initialization was avoided. At the same time, the algorithm can evaluates the importance of not only fuzzy rules but also input variables, and so each of input variables and fuzzy rules can be revised based on the erroneous decrement （ERR）. The algorithm can be applied not only to modeling, but also to extracting some meaningful fuzzy rules used to acquire knowledge. Compared with D-FNN and the other same methods, it is clear that the algorithm has better superiority in the aspect of study efficiency and performance. At last, simulation experiments were done according to the actual cases, and the results show that the algorithm is valid and high effective.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2010年第6期1375-1379,共5页
Journal of System Simulation
基金
广东省自然科学基金(9151040701000002)
