Acquisition of Fuzzy Control Rules Using Soft Computing Technique

模糊控制中使用的模糊控制规则通常是由人工操作者通过试错来构建的。但当被控对象较为复杂时，人工构建模糊规则就变得困难。本课题旨在利用软计算技术对前、后两部分模糊规则进行调优，自动获取模糊控制规则。提出了“模糊单态推理法”、“模糊泛函推理法”和“加权泛函推理法”等新的模糊控制推理方法，并研究了它们之间的关系。利用神经学中最深层的相关方法和模糊聚类方法，给出了模糊单例推理方法模糊控制规则的先行部分的中心和宽度以及后向部分的位置和权重的调整方法。通过对几个函数的计算机仿真表明，与模糊控制中常用的简化推理方法相比，该方法生成的模糊控制规则数量更少。提出了一种基于计算的两个模糊集之间的距离的距离型模糊推理方法，并利用距离型模糊推理方法讨论了特征空间中的鲁棒联想记忆算法、动态推理算法和反插值算法。我们考虑使用现场可编程门阵列的可进化硬件来运行自主移动机器人Khepera以避免碰撞。树型染色体编码是遗传算法中一种新的染色体编码方式。此外，我们构造了一个四足机器人，并演示了模糊控制下的平衡和位置控制。

Fuzzy control rules used in fuzzy controls are usually constructed through trial and error by human operators. But it becomes difficult to construct the fuzzy rules by humans when the controlled plants are complicated.This project aims to acquire fuzzy control rules automatically by using soft computing techniques in tuning the antecedent and consequent parts fuzzy rules.New fuzzy reasoning methods for fuzzy controls are proposed such as "fuzzy singleton-type reasoning method", "fuzzy functional reasoning method" and "weighted functional reasoning method", and their relationships are investigated.We give a method of tuning the center and width of the antecedent part as well as the position and weight of the consequent part of fuzzy control rules of fuzzy singleton-type reasoning method by using the deepest decendent method used in the neuro technique and fuzzy clustering method. It is shown from the computer simulation of several functions that this method can generate smaller number of fuzzy control rules compared with the case by the simplified reasoning method which is used usually and widely in fuzzy controls.We propose a distance-type fuzzy reasoning method based on the calculated distance between two fuzzy sets, and discuss robust associative memory algorithm, dynamic reasoning algorithm and unterpolation algorithm in the characteristic space using distant-type fuzzy reasoning method.We consider evolvable hardware using field programmable gate array to run autonomous mobile robot Khepera for collision avoidance. As a new type of chromosome coding in genetic algorithm, tree structure-type chromosome coding is proposed. Moreover, we construct a robot with 4 legs and demonstrate the balance and position controls under fuzzy controls.