蚁群算法外文翻译

本科毕业设计(论文) 外文翻译（附外文原文）

学院：信息科学与工程学院

课题名称：蚁群优化算法研究

专业(方向)：网络工程

班级：网络工程10-1班

学生：梁绍仲

指导教师：董明刚

日期： 2014年4月6日

蚁群算法外文翻译

中文：

起源

蚁群算法(ant colony optimization, ACO)，又称蚂蚁算法，是一种用来在图中寻找优化路径的机率型算法。它由Marco Dorigo于1992年在他的博士论文中提出，其灵感来源于蚂蚁在寻找食物过程中发现路径的行为。蚁群算法是一种模拟进化算法,初步的研究表明该算法具有许多优良的性质.针对PID控制器参数优化设计问题,将蚁群算法设计的结果与遗传算法设计的结果进行了比较,数值仿真结果表明,蚁群算法具有一种新的模拟进化优化方法的有效性和应用价值。

原理

各个蚂蚁在没有事先告诉他们食物在什么地方的前提下开始寻找食物。当一只找到食物以后，它会向环境释放一种信息素，吸引其他的蚂蚁过来，这样越来越多的蚂蚁会找到食物！有些蚂蚁并没有象其它蚂蚁一样总重复同样的路，他们会另辟蹊径，如果令开辟的道路比原来的其他道路更短，那么，渐渐地更多的蚂蚁被吸引到这条较短的路上来。最后，经过一段时间运行，可能会出现一条最短的路径被大多数蚂蚁重复着。

为什么小小的蚂蚁能够找到食物？他们具有智能么？设想，如果我们要为蚂蚁设计一个人工智能的程序，那么这个程序要多么复杂呢？首先，你要让蚂蚁能够避开障碍物，就必须根据适当的地形给它编进指令让他们能够巧妙的避开障碍物，其次，要让蚂蚁找到食物，就需要让他们遍历空间上的所有点；再次，如果要让蚂蚁找到最短的路径，那么需要计算所有可能的路径并且比较它们的大小，而且更重要的是，你要小心翼翼的编程，因为程序的错误也许会让你前功尽弃。这是多么不可思议的程序！太复杂了，恐怕没人能够完成这样繁琐冗余的程序。

然而，事实并没有你想得那么复杂，上面这个程序每个蚂蚁的核心程序编码不过100多行！为什么这么简单的程序会让蚂蚁干这样复杂的事情？答案是：简单规则的涌现。事实上，每只蚂蚁并不是像我们想象的需要知道整个世界的信息，他们其实只关心很小范围内的眼前信息，而且根据这些局部信息利用几条简单的规则进行决策，这样在蚁群这个集体里，复杂性的行为就会凸现出来。这就是人工生命、复杂性科学解释的规律！那么，这些简单规则是什么呢？

1、范围：

蚂蚁观察到的范围是一个方格世界，蚂蚁有一个参数为速度半径（一般是3），那么它能观察到的范围就是3*3个方格世界，并且能移动的距离也在这个范围之内。

2、环境：

蚂蚁所在的环境是一个虚拟的世界，其中有障碍物，有别的蚂蚁，还有信息素，信息素有两种，一种是找到食物的蚂蚁洒下的食物信息素，一种是找到窝的蚂蚁洒下的窝的信息素。每个蚂蚁都仅仅能感知它范围内的环境信息。环境以一定的速率让信息素消失。

3、觅食规则：

在每只蚂蚁能感知的范围内寻找是否有食物，如果有就直接过去。否则看是否有信息素，并且比较在能感知的范围内哪一点的信息素最多，这样，它就朝信息素多的地方走，并且每只蚂蚁都会以小概率犯错误，从而并不是往信息素最多的点移动。蚂蚁找窝的规则

和上面一样，只不过它对窝的信息素做出反应，而对食物信息素没反应。

4、移动规则：

每只蚂蚁都朝向信息素最多的方向移，并且，当周围没有信息素指引的时候，蚂蚁会按照自己原来运动的方向惯性的运动下去，并且，在运动的方向有一个随机的小的扰动。为了防止蚂蚁原地转圈，它会记住最近刚走过了哪些点，如果发现要走的下一点已经在最近走过了，它就会尽量避开。

5、避障规则：

如果蚂蚁要移动的方向有障碍物挡住，它会随机的选择另一个方向，并且有信息素指引的话，它会按照觅食的规则行为。

6、播撒信息素规则：

每只蚂蚁在刚找到食物或者窝的时候撒发的信息素最多，并随着它走远的距离，播撒的信息素越来越少。

根据这几条规则，蚂蚁之间并没有直接的关系，但是每只蚂蚁都和环境发生交互，而通过信息素这个纽带，实际上把各个蚂蚁之间关联起来了。比如，当一只蚂蚁找到了食物，它并没有直接告诉其它蚂蚁这儿有食物，而是向环境播撒信息素，当其它的蚂蚁经过它附近的时候，就会感觉到信息素的存在，进而根据信息素的指引找到了食物。

问题：

说了这么多，蚂蚁究竟是怎么找到食物的呢？在没有蚂蚁找到食物的时候，环境没有有用的信息素，那么蚂蚁为什么会相对有效的找到食物呢？这要归功于蚂蚁的移动规则，尤其是在没有信息素时候的移动规则。首先，它要能尽量保持某种惯性，这样使得蚂蚁尽量向前方移动（开始，这个前方是随机固定的一个方向），而不是原地无谓的打转或者震动；其次，蚂蚁要有一定的随机性，虽然有了固定的方向，但它也不能像粒子一样直线运动下去，而是有一个随机的干扰。这样就使得蚂蚁运动起来具有了一定的目的性，尽量保持原来的方向，但又有新的试探，尤其当碰到障碍物的时候它会立即改变方向，这可以看成一种选择的过程，也就是环境的障碍物让蚂蚁的某个方向正确，而其他方向则不对。这就解释了为什么单个蚂蚁在复杂的诸如迷宫的地图中仍然能找到隐蔽得很好的食物。

当然，在有一只蚂蚁找到了食物的时候，大部分蚂蚁会沿着信息素很快找到食物的。但不排除会出现这样的情况：在最初的时候，一部分蚂蚁通过随机选择了同一条路径，随着这条路径上蚂蚁释放的信息素越来越多，更多的蚂蚁也选择这条路径，但这条路径并不是最优（即最短）的，所以，导致了迭代次数完成后，蚂蚁找到的不是最优解，而是次优解，这种情况下的结果可能对实际应用的意义就不大了。

蚂蚁如何找到最短路径的？这一是要归功于信息素，另外要归功于环境，具体说是计算机时钟。信息素多的地方显然经过这里的蚂蚁会多，因而会有更多的蚂蚁聚集过来。假设有两条路从窝通向食物，开始的时候，走这两条路的蚂蚁数量同样多（或者较长的路上蚂蚁多，这也无关紧要）。当蚂蚁沿着一条路到达终点以后会马上返回来，这样，短的路蚂蚁来回一次的时间就短，这也意味着重复的频率就快，因而在单位时间里走过的蚂蚁数目就多，洒下的信息素自然也会多，自然会有更多的蚂蚁被吸引过来，从而洒下更多的信息素……；而长的路正相反，因此，越来越多地蚂蚁聚集到较短的路径上来，最短的路径就近似找到了。也许有人会问局部最短路径和全局最短路的问题，实际上蚂蚁逐渐接近全局最短路的，为什么呢？这源于蚂蚁会犯错误，也就是它会按照一定的概率不往信息素高

的地方走而另辟蹊径，这可以理解为一种创新，这种创新如果能缩短路途，那么根据刚才叙述的原理，更多的蚂蚁会被吸引过来。

引申

跟着蚂蚁的踪迹，你找到了什么？通过上面的原理叙述和实际操作，我们不难发现蚂蚁之所以具有智能行为，完全归功于它的简单行为规则，而这些规则综合起来具有下面两个方面的特点：

1、多样性

2、正反馈

多样性保证了蚂蚁在觅食的时候不置走进死胡同而无限循环，正反馈机制则保证了相对优良的信息能够被保存下来。我们可以把多样性看成是一种创造能力，而正反馈是一种学习强化能力。正反馈的力量也可以比喻成权威的意见，而多样性是打破权威体现的创造性，正是这两点小心翼翼的巧妙结合才使得智能行为涌现出来了。

引申来讲，大自然的进化，社会的进步、人类的创新实际上都离不开这两样东西，多样性保证了系统的创新能力，正反馈保证了优良特性能够得到强化，两者要恰到好处的结合。如果多样性过剩，也就是系统过于活跃，这相当于蚂蚁会过多的随机运动，它就会陷入混沌状态；而相反，多样性不够，正反馈机制过强，那么系统就好比一潭死水。这在蚁群中来讲就表现为，蚂蚁的行为过于僵硬，当环境变化了，蚂蚁群仍然不能适当的调整。

既然复杂性、智能行为是根据底层规则涌现的，既然底层规则具有多样性和正反馈特点，那么也许你会问这些规则是哪里来的？多样性和正反馈又是哪里来的？我本人的意见：规则来源于大自然的进化。而大自然的进化根据刚才讲的也体现为多样性和正反馈的巧妙结合。而这样的巧妙结合又是为什么呢？为什么在你眼前呈现的世界是如此栩栩如生呢？答案在于环境造就了这一切，之所以你看到栩栩如生的世界，是因为那些不能够适应环境的多样性与正反馈的结合都已经死掉了，被环境淘汰了！

蚁群算法的特点

1) 蚁群算法是一种自组织的算法。在系统论中，自组织和它组织是组织的两个基本分类，其区别在于组织力或组织指令是来自于系统的内部还是来自于系统的外部，来自于系统内部的是自组织，来自于系统外部的是他组织。如果系统在获得空间的、时间的或者功能结构的过程中，没有外界的特定干预，我们便说系统是自组织的。在抽象意义上讲，自组织就是在没有外界作用下使得系统墒增加的过程(即是系统从无序到有序的变化过程)。蚁群算法充分休现了这个过程，以蚂蚁群体优化为例子说明。当算法开始的初期，单个的人工蚂蚁无序的寻找解，算法经过一段时间的演化，人工蚂蚁间通过信息激素的作用，自发的越来越趋向于寻找到接近最优解的一些解，这就是一个无序到有序的过程。

2) 蚁群算法是一种本质上并行的算法。每只蚂蚁搜索的过程彼此独立，仅通过信息激素进行通信。所以蚁群算法则可以看作是一个分布式的多agent系统，它在问题空间的多点同时开始进行独立的解搜索，不仅增加了算法的可靠性，也使得算法具有较强的全局搜索能力。

3) 蚁群算法是一种正反馈的算法。从真实蚂蚁的觅食过程中我们不难看出，蚂蚁能够最终找到最短路径，直接依赖于最短路径上信息激素的堆积，而信息激素的堆积却是一个正反馈的过程。对蚁群算法来说，初始时刻在环境中存在完全相同的信息激素，给予系统一个微小扰动，使得各个边上的轨迹浓度不相同，蚂蚁构造的解就存在了优劣，算法采用的

反馈方式是在较优的解经过的路径留下更多的信息激素，而更多的信息激素又吸引了更多的蚂蚁，这个正反馈的过程使得初始的不同得到不断的扩大，同时又引导整个系统向最优解的方向进化。因此，

正反馈是蚂蚁算法的重要特征，它使得算法演化过程得以进行。

（4) 蚁群算法具有较强的鲁棒性。相对于其它算法，蚁群算法对初始路线要求不高，即蚁群算法的求解结果不依赖子初始路线的选择，而且在搜索过程中不需要进行人工的调整。其次，蚁群算法的参数数目少，设置简单，易于蚁群算法应用到其它组合优化问题的求解。

蚁群优化算法最初用于解决TSP问题，经过多年的发展，已经陆续渗透到其他领域中，如，图着色问题、大规模集成电路设计、通讯网络中的路由问题以及负载平衡问题、车辆调度问题等。蚁群算法在若干领域己获得成功的应用，其中最成功的是在组合优化问题中的应用。

在网络路由处理中，网络的流量分布不断变化，网络链路或结点也会随机地失效或重新加入。蚁群的自身催化与正向反馈机制正好符合了这类问题的求解特点，因而，蚁群算法在网络领域得到一定应用。蚁群觅食行为所呈现出的并行与分布特性使得算法特别适合于并行化处理。因而，实现算法的并行化执行对于大量复杂的实际应用问题的求解来说是极具潜力的。

在某群体中若存在众多无智能的个体，它们通过相互之间的简单合作所表现出来的智能行为即称为集群智能(Swarm Intelligence)。互联网上的交流，不过是更多的神经元连接(人脑)通过互联网相互作用的结果，光缆和路由器不过是轴突和突触的延伸。从自组织现象的角度上看，人脑的智能和蚁群也没有本质上的区别，单个神经元没有智能可言，单个蚂蚁也没有，但是通过连接形成的体系，是一个智能体。

英文：

Origin

Ant colony algorithm (ant colony optimization, ACO), also known as ant algorithm is a probability -based algorithm is used to find the optimal path in the graph . It in his doctoral thesis proposed by Marco Dorigo in 1992 , inspired by the behavior of the path of the ants found in the process of looking for food . Ant colony algorithm is a simulated evolutionary algorithm , preliminary studies show that the algorithm has many excellent properties. Optimization design for PID controller parameters , the results of the results of the ant colony algorithm design and genetic algorithm are compared , the numerical simulation the results show that ant colony algorithm is a new kind of simulated evolutionary optimization effectiveness and value.

Principle

Individual ants do not tell them in advance what the food in the context of local start looking for food. When one find food after it is released into the environment of a pheromone to attract other ants to come , so the ants will find more and more food ! Some ants and other ant not as total repeating the same way, they will open a way , so that if the open road original shorter than the other way , then gradually more ants are attracted to a shorter path to this . Finally , after a period of time to run , there may be a shortest path is repeated most of the ants.

Why tiny ants to find food ? They have the intelligence it? Imagine if we want to design an artificial ant program , then this program how complicated you want it? First, you have to let the ants can avoid obstacles , we must give it compiled into the terrain under the appropriate instructions so that they can cleverly avoiding obstacles , and secondly, to make the ants find food , they need to get them to traverse space on all points ; again, if you want the ants to find the shortest path , then you need to calculate all possible paths and compare their size, but more importantly, you have to careful programming , because the wrong procedure may make you come to naught. This is such an incredible program ! Too complicated , I am afraid that no one can accomplish this tedious redundancy program .

However , the truth is not as complicated as you think , at the core of this program , but the program code for each of more than 100 lines of ants ! Why make such a simple procedure ants do such complicated things ? The answer is: the emergence of simple rules. In fact, each ant is not as we imagine the whole world need to know the information , they are actually only interested in the immediate

information within a small range , and make decisions based on partial information on the use of these few simple rules , so in this colony collectively , the complexity of behavior will stand out . This is artificial life , complexity science to explain the law ! So, what are these simple rules be?

1 range:

Ants observed range is a square in the world , there is an argument for the speed of ants radius ( usually 3 ) , then it is observed in the range of 3 * 3 grid world, and is also able to move from this range inside .

2 Environment:

Ants where the environment is a virtual world , where there are obstacles , there are other ants , as well as pheromones, information known two types of ants to find food food shed pheromones , one is to find the nest ant nest shed pheromone . Each ant is only able to perceive environmental information within its range. Environment at a rate to allow pheromone disappears.

3 foraging rules:

Looking at the range of each ant can sense whether there is food , if there is direct in the past. Otherwise, see if there are pheromones , and compare within the scope of what point can perceive pheromones at most , so it toward the pheromone many places to go, and the ants are a small probability of making mistakes , which is not to information Su moved up to a point . Ant nest to find the rules and the same as above , but its nest react to pheromones , pheromone and food did not respond .

NOTE 3: This small probability of mistakes is critical, ie ant pheromones have less likely to move toward the place , which makes the ants have the opportunity to develop new path , thereby making it possible to find a shorter path than the current path. The mistake in the algorithm implemented by the wheel choice .

4 mobile rules

Each ant pheromones direction toward shifting up , and when no pheromone guidelines around when the ants in the direction of movement of the inertia of their original motion continues , and there is a small random perturbations in the direction of movement. To prevent ants standing around , it will remember what has recently gone through a point , if we find the next point to go has gone through recently, it will try to avoid .

5 obstacle avoidance rules

If you ant to be moving in the direction obstructions blocking , it will randomly choose another direction, and there are pheromones guidelines , then it will follow the rules of foraging behavior .

6 sowing pheromone rules

Each ant nest at just the time to find food or pheromones send up to Caesar , and with it go far away , sow less pheromone .

According to several of these rules, and there is no direct relationship between the ants , but each ant and environmental interaction occurs , and through this link pheromone , in fact, the correlation between individual ants up . For example, when an ant find food , it does not directly tell the other ants have food here , but to spread the pheromone environment , while other ants pass near it, will feel the presence of pheromones , then according to guidelines pheromone found food.

Problem:

Having said that , what is how ants find food for it? In the absence of ants find food when the environment is not useful pheromone , then why would a relatively effective ants find food? This is attributed to the ants move rules , especially in the absence of pheromone when moving rules. First, it can be possible to maintain a certain inertia , such that the ants moved forward as far as possible ( starting direction of the front of a stationary random ) , and not place unnecessary vibration or spin ; secondly , have a certain randomness ant , even with a fixed direction, but it can not go on like particles as linear motion , but there is a random disturbance . This makes the ants in motion with a certain purpose, try to keep the original direction , but there are new temptations , especially when it comes to an obstacle when it will instantly change direction, which can be seen as a process of selection , but also environment so that an obstacle ant right direction , while the other direction is not right. This explains why a single ant on the map , such as complex maze of hidden still be able to find very good food .

Of course, there is an ant to find food when most ants will soon find food along the pheromone . But do not rule out such a situation : In the beginning, part of the ants by randomly selecting the same path , this path as the ant pheromone release more and more, more ants choose this path , but this is not the best path ( i.e., the shortest) , and therefore , resulted in the completion of iterations , the ants found is not the optimal solution, but sub-optimal solution in this case may be the result of the practical significance of the insignificant.

Ants how to find the shortest path ? This is thanks to the pheromone , the other thanks to the environment , specifically the computer clock . Pheromone many places where ants apparently after many ants gathered thus there will be more to come . Suppose there are two paths leading to food from the nest , the beginning, the number of ants walking two roads as much ( or more than a long way ants , it does not matter ) . When the ants along a road to reach the terminal will immediately returned, so that a short round-trip one -way ants short of time , this also means that the frequency of repeated soon , and thus the number of ants in the unit of time , the more traveled ,

pheromones naturally shed more naturally have more ants are attracted, thus shed more pheromones ...... ; rather a long way on the contrary, therefore , more and more ants gathered over the short path up, shortest path found on the approximation . Maybe someone will ask questions of local and global shortest path shortest , in fact, the ants gradually approaching global shortest , and why ? This stems from the ants will make mistakes , that is, it will not go to high places pheromone and another way to go according to a certain probability , which can be understood as a kind of innovations that if we can shorten the road , then in accordance with the principles just described , more ants are attracted.

Extended

Follow the trail of ants you find what ? Through narrative and practice the principles of the above , we find that the reason why the ants with intelligent behavior , entirely due to its simple rules of behavior , and these rules together with the characteristics of the following two areas:

1,diversity

2,the positive feedback

Diversity ensures that when foraging ants did not mention a dead end and an infinite loop , positive feedback mechanism will ensure a relatively good information can be saved. We can diversity be seen as a creative ability, and positive feedback is an enhanced ability to learn . Positive feedback can also be likened to the power of authoritative opinion , but diversity is reflected in the creative authority to break , it is the unique combination of these two makes intelligent behavior cautiously emerged out.

Extended in terms of evolution, social progress, human nature is actually innovation are inseparable from these two things , the diversity of the system to ensure the ability to innovate, to ensure that the excellent characteristics of positive feedback can be strengthened, both to be just right combined . If diversity surplus , which is the system is too active, which is equivalent to the random movement of ants too much , it will fall into chaos ; By contrast , diversity is not enough positive feedback mechanism is too strong , then the system is like a backwater. In terms of this colony on the performance of the ant 's behavior is too stiff , when the environment changes , the ants are still not properly adjusted .

Given the complexity of the underlying intelligent behavior is based on the emergence of the rule , since the underlying rules of the diversity and characteristics of positive feedback , then maybe you will ask where these rules come from? Diversity and positive feedback is where they come from ? My own opinion : the rules of nature and evolution . The evolutionary nature of the basis of the speakers also reflected in the diversity and unique combination of positive feedback . This ingenious combination is why? Why in the world presented in front of you is so lifelike it? The answer lies in the environment created all of this,

the reason why you see the life of the world , because the combination of positive feedback and those who can not adapt to the diversity of the environment are already dead, the environment has to be eliminated !

Characteristics of ant colony algorithm

（1) ant colony algorithm is a self-organizing algorithms. In systems theory, self- organization and its organization is the organization 's two basic categories , the difference is that the organization is an internal force or organizational instruction from the system or from outside the system, from within the system is self-organization, from the his organization is external to the system . If the system or to obtain a functional structure of the process space of time, there is no specific interventions outside , we say the system is self- organized . In an abstract sense, since no outside organization is in effect making the system entropy increases under the process ( that is, the system change process from disorder to order ) . Ant colony algorithm is now fully off this process to ant colony optimization examples. When the algorithm starts early , to find a single solution disorderly artificial ants , the algorithm through evolution over time, between artificial ants through the action of hormones , spontaneous growing tendency to find some solution close to the optimal solution , which is a disorder to order process.

（2) ant colony algorithm is an inherently parallel algorithms. Each ant search process independent of each other and only communicate through pheromones . So the ant colony algorithm can be seen as a distributed multi-agent system , it also began to conduct independent multi-point solution to search the problem space , not only increases the reliability of the algorithm, but also makes the algorithm has strong global search capability .

（3） ant colony algorithm is an algorithm positive feedback . Real ants from foraging process , we can easily see that the ants can finally find the shortest path , the shortest path is directly dependent on the accumulation of pheromones , hormones and the accumulation of information but it is a process of positive feedback . Ant colony algorithm , the initial time the same pheromones exist in the environment , giving the system a tiny perturbations , making each side of the track is not the same concentration , the solution structure of ants on the merits of the existence of a feedback manner algorithm used after the solution is in the optimum path leaving more pheromones , hormones and more information and attract more ants, the positive feedback process allows different initial constantly expanding , while guiding the whole system evolution toward the optimal solution . Therefore ,

Positive feedback is an important feature of ant algorithm , which makes the algorithm evolution to proceed.

(4)ant colony algorithm has strong robustness. Compared with other algorithms ,

ant colony algorithm for the initial route less demanding that the results of the ant colony algorithm for solving the child does not depend on the initial route selection , but also in the search process does not require manual adjustments. Second, the small number of parameters ant colony algorithm , set up a simple , easy-to- colony algorithm is applied to solve other combinatorial optimization problems .

Ant colony optimization algorithm for solving TSP problem initially , after years of development, has gradually penetrated into other areas , such as graph coloring problem, LSI design , communication network routing problems , and load balancing issues , vehicle scheduling problem and so on. Ant colony algorithm has been a success in several areas of application , which is the most successful application of combinatorial optimization problems.

In the network routing process, changing the distribution of network traffic , network link or node failure or will randomly re- join. Autocatalytic and positive feedback mechanisms colony fits the characteristics of this kind of problem solving , and therefore , the ant colony algorithm applied in a certain network area . Parallel and distributed characteristics exhibited by ant foraging behavior makes the algorithm is particularly suitable for parallel processing . Thus, the parallel execution of algorithms for a large number of practical applications to solve complex problems is highly promising .

In a group , if there are many non- intelligent individuals , they cooperate with each other by simply demonstrated between intelligent behavior is called swarm intelligence (Swarm Intelligence). Exchanges on the Internet , but is more neuronal connections ( the brain ) results , cable and router interaction via the Internet , but is an extension of axons and synapses . From the point of view of self-organization , intelligence and ant brain is also no difference in essence , a single neuron is no intelligence at all, there is no single ant , but the system is connected via the formation of an agent .