读书笔记----fuzzy control and identification 1
introduction
最近在做一个非线性控制系统的项目,搞得很崩溃,读一本fuzzy control[1]的书,然后打算用matlab fuzzy logic toolbox实践一下。
book content
chapter1: introduction
fuzzy system
作者并没有给出fuzzy system的定义,强调人在决策时,根据rules,在某种情形下,平衡各种情况,最终给出决策。
You have a set of rules
(probably unspoken) in your mind that help you decide which route to take
fuzzy system的优势:
1)适应很难建立数学模型的场景,小男孩手顶扫帚
2)具备一定的鲁棒性,即使扫帚收到扰动,小男孩还是能顶住
3)已经在航天领域应用了部分模糊控制的方法,虽然传统算法可以通过稳定性分析,但是这种分析是建立在模型准确和已知的条件下,实际的模型可能有偏差,同时在飞机起飞场景下,采用模糊算法的人类驾驶员仍然是主导。
expert experience
专家经验是模糊控制中至关重要的内容,expert 通过自身practice,总结出experience。
何时使用fuzzy control
- 对于线性时不变系统,用fuzzy control是多此一举。
- 针对复杂、非线性、时变系统
control
-
control objective
stablization, tracking, and model following
stablization: 系统稳定:尤其是开环系统本身不稳定,要求闭环系统稳定
tracking:被控对象的运动轨迹按照设定的轨迹运行:强调位置上满足设计需求
model following: 被控对象的动力学特性符合设计需求:强调时间上满足设计需求
interconnection of several subsystem
使用传递框图,表示控制系统,之间系统之间的关系。
system identification and adaptive control
- fuzzy system identification
不同于线性系统的系统辨识,不需要在一个特定模型下做参数辨识。 - adaptive fuzzy control
Adaptive control is a method by which the system behavior is monitored online in
real time and the control continually updated and adjusted to adapt to uncertainties or
changes in the plant.
实时的观测系统行为,控制器连续的更新和修改参数以适应模型不确定性和模型的变化。
- indirected adaptive control
在线辨识模型,根据模型修改控制参数 - directed adaptive control
不需要通过系统辨识,控制参数直接改变。
总结
- fuzzy logic and expert experience
fuzzy control 是模仿人的决策过程,核心是专家经验。 - 使用范围
computer cost, 适用于复杂系统 - 关注点
stablization, tracking and model following
chapter 2: basic concepts of fuzzy sets
fuzzy set
- 理解fuzzy set
a fuzzy set is a collection of real numbers having
partial membership in the set
这句话比较难理解,作为对比的内容,number可以partially belong 一个fuzzy set,但是不能partially belong一个conventional 的集合:如个子高的人和高于1.8m的人。(猜测fuzzy control的数学基础有一部分是概率论)
This is in contrast with conventional, or crisp sets, to
which a number can belong or not belong, but not partially belong
- 总结fuzzy set
1.集合成员
2.成员属于这个集合的程度【0,1】,0完全不是,1完全是,(0,1)部分符合。 - lingusitic variable
“height"是lingusitc variable
"h"是math symbol
这些变量的特点都是抽象,不要精确的定量 - universe of discourse
往往是R(全体实数),但是因为很多数的属于set的degree显然是0,
因而有一个effective universe of discourse:高度:[0, +∞), 人的高度:[1,3) -
a variable universe of discourse is ,变量x满足(),
member function能反映x属于fuzzy set的程度。
- 常见fuzzy set
以下各种形状都是可以的,关键是形状背后的物理意义,例如图4,用它表示“温暖”就很奇怪。
1.三角函数member function
2.高斯函数的member function
3.singleton shaped member function
4.arbitary shaped member function
useful concepts for fuzzy set
- support
membership function 大于0的成员。 - α-cut
membership function 大于α的成员 - height
peak of membership function, 往往是1 - normal fuzzy set
在universe of discourse 中,至少有一个点的membership function 是1 - convex fuzzy set
membership function 并不一定要求是convex function,并且member function即使是convex function也不一定是convex fuzzy set
fuzzy set 理论计算
- fuzzy subset
CRT是WARM的subset
Let M1 and M 2 be two fuzzy sets defined for a variable on the universe of
discourse X, and let their associated membership functions be μ1(x) and
μ2(x), respectively. Then, M1 is a fuzzy subset of M 2 (or M1 ⊆ M 2) if
μ1(x) ≤ μ2(x)∀x ∈ X
- Fuzzy Compliment
Consider a fuzzy set M defined for a variable on the universe of discourse X,
and let M have associated membership function μM(x). The fuzzy compliment of M is a fuzzy set ¯ M characterized by membership function
μM ¯(x) = 1 − μM(x).
- Fuzzy Intersection (AND)
Let M1 and M 2 be two fuzzy sets defi ned for a variable on the universe
of discourse X, and let their associated membership functions be
为T-norm
- fuzzy union(or)
为T-conorm
- Fuzzy Cartesian Product
summary
人在做决策时,例如给消费,将服务转换成fuzzy set, good service, bad service, normal service, 然后决定要给的消费范围,nice tip, normal tip, little tip, 最后要将决策转换成crisp number, 例如10美元。
所以这章的内容非常重要,在后面解决问题时,我们需要将crisp number转换成fuzzy set,然后通过fuzzy logic 作出决策,最后再转换成crisp number。
references
[1] Lilly, John H. Fuzzy control and identification. John Wiley & Sons, 2011.