【作者】 张华；

【导师】 王宝文；

【作者基本信息】 燕山大学 ， 计算机应用技术， 2007， 硕士

【摘要】 稀疏规则库广泛的存在于模糊推理系统中。当出现稀疏模糊规则库时,采用传统的模糊推理方法是得不到任何结果的。由此,产生了模糊插值推理方法。目前,已经存有很多的插值推理方法,但他们一般计算复杂,且不能保证结果的正规凸性。为此,本文的目标就是对原有的稀疏模糊插值推理方法进行改进,并且提出更好的稀疏模糊插值推理方法。首先,本文分析了Zhiheng Huang和Qiang Shen提出的基于重心的模糊插值推理方法,但该方法只适用于三角形隶属函数。针对这个缺点,本文改进了基于重心的模糊插值推理方法,使其不但适用于三角形隶属函数,同时也适用于梯形隶属函数。其次,通过仔细的分析研究和实例仿真,发现Y.Shi和M.Mizutomo提出的模糊拉格朗日插值推理方法不能保证最后结果的正规凸性。因而,本文在拉格朗日插值函数的基础上给出了基于几何参数的模糊拉格朗日插值推理方法。该方法不仅能保证最后结果的正规凸性,而且能适用于其他的正规凸模糊集。最后,给出了一种基于核集的稀疏模糊规则库条件下的模糊推理方法。该方法计算简单,而且当模糊规则库是稀疏型时,利用该方法得到的模糊推理结论的隶属函数是正规的凸模糊集。并把给出的基于核集的模糊插值推理方法从一维情况扩展到多维情况,拓宽了该方法的应用范围。更多还原

【Abstract】 When rule base is sparse, any reasoning result can not be gotten by traditional CRI method, however interpolative reasoning methods can tackle this problem. Now, there are some interpolative reasoning methods, but they may include complex computation and can’t keep the convexity of the reasoning consequence. So, the target of this dissertation is the research on how to improve the customary reasoning, and proposed a new sparse fuzzy reasoning method.Firstly, the interpolative reasoning method based on center of Gravity was described. But, this method didn’t allow the conditions appearing in the antecedent part of rule to be represented by trapezoidal fuzzy numbers, so this paper generalized the method. The generalized fuzzy interpolative method based on center of gravity allows the conditions appearing in the antecedent part and the consequence of the rules to be represented not only by triangular fuzzy numbers but also by trapezoidal fuzzy numbers.Secondly, by analyzed and simulated, it was found that the fuzzy lagrange’s interpolative reasoning method can’t guarantee the reasoning consequences are normal and convex. So a fuzzy lagrange’s interpolative method based on geometric parameter was presented. Reasoning is simple by the method; moreover it can keep the convexity of the reasoning consequence.Finally, a fuzzy interpolative reasoning method based on core was proposed. When the fuzzy rule bases are sparse, the reasoning conclusions can be derived simply and are normal convex fuzzy sets. By this new method the new reasoning method is programmed in MATLAB subsequently. The one dimension sparse fuzzy reasoning method was expanded into multi-dimension fuzzy reasoning, and its application area was spread.更多还原