关于模糊逻辑（FL）和神经网络（NN）。

PEN

我研究了模糊逻辑和神经网络。我认为神经网络比模糊逻辑更有前途，但它需要太多的处理能力来实现便宜。即使是模糊逻辑也需要太多的处理能力。

我见过许多有关使用模糊逻辑进行液压运动控制的学生和研究文档。到目前为止，我还没有发现它们中的任何一个都很好。即使德州仪器将模糊逻辑控制与PID控制进行比较的文档也很差。很容易看出作者不理解控制理论，因此他们无法正确优化PID增益。作者使PID控制看起来不好，所以模糊逻辑控制看起来比较好。

我会诋毁作者撰写诸如真相和误导读者等想要学习的垃圾。

一些模糊逻辑论文希望minimin一个PID。如果是这样，模糊逻辑怎么会更好？

一些模糊逻辑论文希望使用模糊逻辑来动态调整PID增益。
这些论文比较好，但PID不是阻尼液压缸下的最佳控制形式。模糊逻辑论文忽略前馈。

模糊逻辑和神经网络优于简单的PID的优点是它们可以有两个输出而不是一个，但是由于伺服液压控制只需要一个输出到伺服阀来控制位置，速度和加速度，我没有看到任何需要为模糊逻辑

I have researched fuzzy logic and neural nets.   I think neural nets are more promising than fuzzy logic but it takes too much processing power to implement cheaply.   Even fuzzy logic takes too much processing power..

I have seen many student and research documents about using fuzzy logic for hydraulic motion control. So far I have not found any of them very good.  Even the docment done by Texas Instruments that compare fuzzy logic control to PID control is poor.   It is easy to see that the author do not understand control theory so they couldn't optimize the PID gains correctly.   The authors make the PID control look bad so the fuzzy logic control looks good by comparison.

I would discredit the authors for writing such garbage as truth and misleading reader that want to learn.

Some fuzzy logic papers want to either minmic a PID.  If so then how can fuzzy logic be better?

Some fuzzy logic papers want to use fuzzy logic to adjust the PID gains dynamically.
These papers are better but PID is not the best form of control for under damped hydraulic cylinders.  The fuzzy logic papers ignore feedforwards.

The advantage that fuzzy logic and neural nets have over a simple PID is that they can have two outputs instead of one but since servo hydraulic control only needs one output to the servo valve to contro position, velocity and acceleration I don't see any need for fuzzy logic or neural nets.

I welcome anybody to post a link to pdf files about neural nets and fuzzy logic applied to hydraulic servo control.

What students and professor forget is that in the end the motion can be described as a differential equation..  It may not be a linear differential equation but it will still be a differential equation that can be solve fairly accurately using Runge-Kutta.

蜻蜓

我那个加载题目，好像是阻尼液压，我知道控制中用到的是PID，但我不知道用没用到其他控制算法，问题是频率和幅值难以提高，可以肯定不是液压和机械的问题。我估计是负载难以控制或者是控制算法问题。

数字液压

Fuzzy Logic和Neural Nets针对不同的被控系统体现的效果能力是不同的。模糊逻辑更适合用简单的条件判句式方式针对于不确定或者未知的被控对象进行定性方式的控制，将无法简单描述（建模）或尚未进行描述的被控对象快速实现控制收敛，是一种通过简单方法实现更广泛有效控制的相对廉价高效算法。而神经网络则适合多变量多参数系统，这两种算法针对的系统不同，因此直接比较是不合适的。

Fuzzy Logic其实说起来并不复杂，大多数中国人小时候学习自行车的时候都有体会，兄长或者长辈会简单描述一下骑自行车的要领，车子向哪边倒，骑行者就向哪边转舵，倒的速度越快，转舵的幅度和速度就需要越大。。。然后我们通过不断的骑行摸索，很快就能将这个“杂技”学会，其实这个过程就是模糊逻辑控制。它不需要你对骑自行车的所有过程和所有可能产生的影响因素建模并针对性的写出精准的控制算式，而是使用最简单的几个逻辑，通过不断的优化收敛每次转舵的多少和速度与加速度，最终就能很好的开始骑行，刚刚提到的转舵多少、速度和加速度对应的就是模糊集。其实模糊逻辑很容易，但模糊集往往是关键，能不能给出准确的模糊集是模糊控制的成败或好坏的关键。

神经网络用一个城市交通控制比较容易理解。当一个路口发生交通事故，蝴蝶效应向周边路口蔓延，影响的程度与不同道路的通流能力等诸多因素相关。。。如果想快速恢复交通秩序，需要不仅在发生事故的路口采取措施，同时也要在周边按隶属关系和隶属度不同也要采取调控措施，方可快速有效的改善事故造成的影响。。。智能交通是神经网络控制的经典应用，这仅仅是单层神经网络应用。

总而言之，不同的控制算法有它适应的领域和范围，有些看似简单的算法用在合适的领域所发挥的效果可能比看似复杂的更来得高效和低成本。

PEN

“模糊逻辑更适合用简单的条件判句式方式针对于不确定或者未知的被控对象进行定性方式的控制，将无法简单描述（建模）或尚未进行描述的被控对象快速实现控制收敛，是一种通过简单方法实现更广泛有效控制的相对廉价高效算法“。

1.什么使模糊逻辑更适合于控制未建模的系统？结果通过与PID控制相同的反复试验获得。
2.开环系统不会因为模糊逻辑控制而改变。开环极点和零点不会改变。目标是移动闭环极点和零点以获得合适的响应。我可以计算PID增益和feedfowards。用模糊逻辑你猜很多次。
3.如果没有前馈，PID和模糊逻辑性能不佳。
4.模糊逻辑不像骑自行车。模糊逻辑具有隶属函数和解模糊器。谁知道如何设置这些？如果隶属函数模仿PID增益，那么优势是什么？
5.什么版本的去模糊化效果最好？
6.你如何在模糊逻辑中实现等价的积分器或差分器？这与在PID中如何完成它非常相似，因此优势在哪里？
7.几周前公布的简单PID只需要3次乘法和加法。与模糊逻辑相比，这非常有效。我会增加4个乘法，并增加前馈和额外的增益。这仍然比模糊逻辑更有效。

1. What makes fuzzy logic more suitable for control of unmodeled systems?  The results are obtained by trial and error the same as PID control..
2. The open loop system does not change just because it is being controlled by fuzzy logic.  The open loop poles and zeros do not change.  The goal is to move the closed loop poles and zeros to get a suitable response.  I can calculate PID gains and feedfowards.  With fuzzy logic you guess many times.
3. Both PID and fuzzy logic perform poorly without feed forwards.
4. Fuzzly logic is not like riding a bike.  Fuzzy logic has membership functions and defuzzifiers. Who know how to set those up?   If the membership functions mimic PID gains then what is the advantage?
5. What version of defuzzifying is best?
6. How do you implement the equivalent of a integrator or differentiator in fuzzly logic?  It is very similar to how it is done in a PID so where is the advantage?
7. The simple PID that was posted a few weeks ago required only 3 multiply and adds.  That was very efficient compared to fuzzy logic.  I would add 4 more multiply and  adds for feedforwards and an additional gain.  This is still much more efficient than fuzzy logic.

PEN

又一个模糊逻辑与PID文件。
http://deltamotion.com/peter/pdf ... 10e392a127e5d65.pdf
请看公式3下面的段落。作者指出了PID控制的一个主要问题。看看我上面的pdf。我对集成商没有任何问题。我会让这些学生失败。更糟糕的是，没有人发表本文IJCSI.org，似乎也理解控制理论。任何人如何信任IJCSI.org发布的内容？

图2中的响应不好。这可能是典型的，但不是最佳的。

代码在131页。ITAE和IAE使用不正确。我曾尝试过这种方法。如果设定值从1变为2，则此方法失败。有一个正确的方式来使用ITAE，但没有在这里显示。

公式10显示了传递函数。只要看看它，我就知道PID不是控制这个系统的最佳方法。

第133页的回应是可疑的。图6显示了类似PID的FLC在不到5秒的时间内从0移动到51弧度。在开环增益之上，Km是1（弧度/秒）/伏特。如果控制器输出10伏，它将以10弧秒的速度移动，但有加速和减速时间。你所看到的是控制器输出大部分时间必须在10伏。

我的观点是，在模糊逻辑与PID方面存在很多误区。考虑到控制输出限于10伏特，很难获得比所示更快的响应。如果移动到10或20弧度，最好不要达到输出限制。

我可以展示更多的FL与PID论文。

Yet another fuzzy logic vs PID document.
http://deltamotion.com/peter/pdf ... 10e392a127e5d65.pdf
Look at the paragraph below formula 3.   The author states a major problem with PID control.   Look at my pdf above.   I have no problems with integrator wind up.  I would fail these students.   What is worse is that no one that published this paper, IJCSI.org , seems to understand control theory either.  How can anybody trust what is published by IJCSI.org?

The response in Fig 2 is not good.  It may be typical but it isn't optimal.

There is code on page 131.   ITAE and IAE are being used incorrectly.  I have tried this method before.  If the set point change from 1 to 2 this method fails.   There is a right way to use ITAE but it is not shown here.

Equation 10 shows the transfer function.  Just by looking at it I know that a PID is not the optimal way to control this system.

The response on page 133 is suspicious.  Fig 6 shows the PID-like FLC moving from 0 to 51 radians in less than 5 seconds.   Above the open loop gain, Km,  is 1 (rad/sec)/volt.   If the controller output 10 volts it would move at 10 radians persecond but there is acceleration and deceleration time.    What you see is the controller output must be at 10 volts most of the time.

My point is that there is a lot of misrepresentation when it comes to fuzzy logic vs PID.   It would be hard to get a faster response than shown given the control output is limited to 10 volts.  It would have been better to move to 10 or 20 radians so the output limit isn't reached.

I have more FL vs PID papers I can show.