ACO算法范文

ACO算法范文（精选3篇）

ACO算法 第1篇

关键词：蜗轮蜗杆,鱼群算法,蚁群算法,故障识别,近邻准则

在蜗杆传动时,由于蜗轮蜗杆摩擦严重,发热大、产生高周疲劳,发生齿轮面点蚀、剥落、及轮齿疲劳折断等故障[1]。可能会带来不可预知的经济损失、人身安全等灾难性后果。因此,蜗轮蜗杆减速器的状态监测与及时故障诊断,不仅能增加减速机的使用周期,减少其维修时间,同时还能降低维修费用和事故发生率。

近年来,国内外学者致力于设备的状态检测与故障诊断技术的研究,将ANN(人工神经网络)等智能算法应用于蜗轮蜗杆的故障识别,取得较好的成果,但现阶段将鱼群及蚁群对蜗轮蜗杆故障识别方法研究较少。即本文对鱼群和蚁群两种算法提出混合改进策略,并利用近邻函数准则的聚类问题将该算法与蜗轮蜗杆故障识别有机地结合策略,寻找出一种基于改进FSA-ACO的故障优化识别处理的诊断方法[2,3,4,5,6,7]。

1 FSA-ACO算法的混合改进

算法初期利用鱼群算法有较快的收敛品质,避免算法初期过早集结到信息素浓度高的路径上,如图1所示。后期引入蚁群算法混合策略,不仅增强了算法的全局寻优和邻域搜索能力,且避免限于局部最优解,实现各算法之间的优缺点互补。如基于TSP问题的图2蚁群算法迭代过程所示。又因两算法遵循的运动规律具有一定的相似性,即人工鱼随着拥挤度浓度高的方向游动,蚂蚁随着信息素浓度高的方向迁移,同时在人工鱼的聚群行为及蚂蚁的更新信息素均取决于伙伴中心或最优个体状态的最优解[3]。综上所述,FSA及ACO算法融合是行之有效的。

将两者算法混合后,利用鱼群的觅食行为,聚群行为和将全局搜索范围缩小。拥挤度与信息素的相似性,在范围缩小后,将其过度到蚁群算法,通过其分布并行式计算和启发式搜索方式,不仅提高收敛速度,还能适当提高收敛的精度,在一定程度上克服局部极值的问题。

2 基于FSA-ACO的故障识别算法

2.1 蜗轮蜗杆的故障识别

本文蜗轮蜗杆是基于振动信号分析与处理的故障识别方法来完成的故障诊断,如图3所示。即通过传感器采集蜗轮蜗杆减速器传动的振动信号,进而对获得的信号经过分析和处理,提取其中与故障类型判别有关的特征向量,再根据所提取得到的故障特征按照某种规律对故障的形态、部位和原因作为样本进行定位、训练和分类,最后构造某种机器学习算法,形成一个诊断网络去识别故障[10,11,12,13]。

基于FSA-ACO混合改进算法的蜗轮蜗杆故障识别可看作一种模式识别,通过测量一些过程参数,然后利用改进的FSA-ACO算法建立一个从测量集合到故障集合的映射关系,从而实现蜗轮蜗杆的故障识别。在蜗轮蜗杆系统发生故障时,其运转与正常状态下是不同征兆的集合,这些不同的征兆对应不同的振动信号特征[2]。

2.2 近邻函数准则

设D={D1,D2,…,DN}为被测系统所有故障的非空有限集合,F={F1,F2,…,FM}为故障集合D引发的所有征兆的非空集合[2]。在蜗轮蜗杆故障诊断中,对故障特征的提取后,征兆集合F中的各个元素与故障集合D是一种复杂的、非比例的对应关系,即其征兆F不能用简单的函数关系或核函数来映射故障D,进而本文引入近邻函数准则策略解决上述故障分类映射问题,同时有效地解决如何在FSA-ACO混合改进的算法与蜗轮蜗杆故障识别联系及融合问题[2,3,4,5,6]。

对于数据集中的任何两个样本yi,yj,若yj是yi的第N个近邻,则称yj对yi的近邻系数为N。若yi是yj的第M个近邻,则称yi对yj的近邻系数为M。当yi和yj互为近邻时则近邻系数为0[2]。因此,近邻函数值τij表示为

在聚类的过程中,规定类内会损失,即当yi和yj被规分同一类时,两者相互“连接”关系,对于每一个这样的连接存在着一个相应的连接损失。这里连接损失规定为这两个样本间的近邻函数值。即总类内损失规定为

当yi和yj被规分不同类时,两者不存在“连接”关系,即τij=0,即总类间损失规定为

式中,C为常数;τtij为其他类间对i类的类间损失,则聚类结果的准则函数

进而蜗轮蜗杆的诊断识别问题转化为求解计算聚类结果的准则函数JNN最小值。

3 构建算法的故障诊断模型

3.1 基于改进算法模型的建立

由故障特征提取的每个蜗轮故障样本数据类似于人工蚂蚁将要访问的地点,而每个人工蚂蚁根据样本数据的之间的信息素的大小及近邻函数,以一定的概率适当选择下一个地点的访问,同时加入自身的数据列表里。每只蚂蚁对全部数据样本均进行一次访问,构成一个有序连接路径,在每个数据列表里,每只蚂蚁的数据列表根据其样本之间的近邻函数值的大小打断近邻函数值较大的两个样本数据的连接,同时计算每一类故障之间的连接损失,形成初步蜗轮蜗杆的故障分类。如此循环类推,寻求样本数据之间的连接损失的最小解为终止条件,进而形成最终的蜗轮蜗杆故障分类结果[2,3]。

3.2 算法具体步骤

通过以上分析,本文算法首先通过故障样本数据建立加权欧氏空间,前期利用鱼群算法寻优特点,淘汰部分与加权距离指标所反映较大的故障样本数据,后期再利用蚁群算法,通过对的最优解求解,进而获得与映射关系相近的蜗轮蜗杆故障征兆,实现对蜗轮蜗杆的故障特征进行分类识别。FSA-ACO混合改进算法具体步骤如下:

步骤1算法初始化,设定种群数、感知距离、初始状态位置、循环最大迭代次数、及拥挤度等参数;

步骤2通过蜗轮蜗杆故障样本数据,建立加权欧氏距离矩阵A,并做近故障样本的邻矩阵M。其中

Pk为故障征兆样本数据中不同特征的加权因子,Mij为样本yi对yj的近邻函数值;

步骤3构建近邻函数矩阵L,其元素为Lij-Mij+Mji-2。在每条“连接关系间”初始化信息素浓度τij(0)=C,C为拥挤度。其目的是种群觅食的向导融合,促使鱼群算法正常过度到蚁群算法;

步骤4种群数K(k=1,2,3,…,m)根据加权欧氏空间里被测数据加权距离指标的对应点,通过鱼群算法对该点的觅食行为、聚群行为,淘汰部分故障样本数据的搜索,其目的是简化后期求JNN的最优解,提高寻优的效率;

步骤5种群数K(k=1,2,3,…,n(n<m))根据故障样本数据的邻近函数矩阵L及初始化信息素浓度τij(0)=C,访问每一故障样本数据的访问地点;

步骤6基于精英策略蚂蚁算法改进策略,将信息素进行如下更新

上式中,Δτ*ij表示精英蚂蚁引起的路径(i,j)上的信息素量的增加;σ是精英蚂蚁的个数;L*是所找出的最优解的路径长度;

步骤7种群按照式

其中,τij(t)为城市i到城市j的信息素浓度;ηij(t)为启发函数;allowK为蚂蚁K待访问城市的集合;α为信息素重要程度因子;β为启发函数重要程度因子。PKij迁移概率选择预计转移路径,并加入到蚂蚁数据列表的故障样本数据加入禁忌表列表后,判断是否所有蚂蚁完成一次访问,否则返回步骤5;

步骤8在数据列表里,根据其样本之间的近邻函数值的大小,打断近邻函数值较大的两个样本数据的连接,同时计算每一类故障之间的连接损失,形成初步蜗轮蜗杆的故障分类;

步骤9重复步骤8操作,直至找到每一只蚂蚁的JNN(k)最小,完成一次故障的识别。取所有蚂蚁中JNN(k)最小的识别结果作为本次循环的求解结果。直至m只蚂蚁都选择这个同一条结果路径或达到指定最大迭代次数为止。

4 算法仿真测试与分析

4.1 试验台振动测试系统

采用蜗轮蜗杆WPA40型号为试验对象,对蜗轮蜗杆减速机进行振动测试,搭建如图4蜗轮蜗杆减速机振动测试系统试验台。该试验台上主要由伺服驱动电机、蜗轮蜗杆减速器、磁粉制动器、振动传感器、扭矩传感器、温度传感器、联轴器、NI采集卡及计算机等组成[10,11,12,13]。

该测试系统中,伺服驱动电机给定输入扭矩6 N·m,转速1 000 rpm,由联轴器连接蜗轮蜗杆减速器的蜗杆,磁粉制动器的加载作为蜗杆传动的负载,结合扭矩传感器测量蜗轮轴的扭矩,并用温度传感器测量蜗轮蜗杆减速机中润滑油的温度。振动传感器通过NI采集卡对减速器的振动信号的采集。再由计算机存储减速器的实时振动信号,给予进一步的时域、频域分析和处理。

在试验时须确保:(1)实验台上的两相连的轴向在一条直线上;(2)装配后需要充分磨合一段时间,同时减速机的润滑油温度达到相对稳定状态后,再进行数据采样和测试;(3)保证运转平稳,各密封处,接合处无漏油,渗油,润滑充分,检查油池和轴承温度;(4)蜗轮齿面初始接触斑点应符合有关标准或设计要求。

4.2 仿真研究

本文基于上述试验对蜗轮蜗杆减速器的振动信号采集,并通过时域、频域分析等方法对其故障特征提取。列出表1不同时域指标与蜗轮故障类型关系。

上述试验均通过伺服驱动电机给定输入扭矩6 N·m,在转速1 000 rpm时,平行于蜗杆的轴向上不同故障类型减速器信号的特征提取,并通过特征提取获得蜗轮蜗杆传动状态征兆与故障数据样本的映射关系[14]。同时再通过上述试验,采集不同故障的蜗杆输入信号数据。

由所得的蜗轮信号输入测量数据样本,通过特征时域指标提取后,加载到本文算法诊断仿真,其中蜗轮蜗杆时域故障指标X1k,X2k…X7k分别为第K个蚂蚁的平均值、均方差、峰值指标、波形因子、峭度指标、裕度指标、脉冲指标的7个指标参数。任意建立一组加权距离指标:(X1k,X2k),(X2k,X4k),(X3k,X6k,X7k),(X3k,X4k,X5k),(X3k,X5k,X6k,X7k)。结合表1不同时域指标与蜗轮故障类型关系,建立故障样本数据。进而利用算法进一步与蜗轮信号输入测量数据样本测试对比。获得如表2所示的指标对故障征兆诊断正确率。

4.3 试验仿真结果

基于上述分析,该算法在不同时域指标与蜗轮故障类型关系作为样本征兆样本数据时,利用不同的蜗轮信号输入时域图去仿真测试,其结果故障征兆诊断识别的正确率均在84%以上,误差在可接受范围内。其发生误差的原因可能是:实验台上拆装时的轴向间的同轴度较难保证一致,蜗轮蜗杆润滑油内部和油温等不确定因素,在建立故障征兆类型关系初期本身会因其特征提取等因素导致误差存在。针对部分因素需要对其进一步优化,本文旨在论述该方法的有效性和可行性,并且误差在可接受范围。

5 结束语

(1)本文通过测试实验将获得的振动信号经过预处理,及时域、频域分析等方法对其故障特征提取,与本文提出的改进鱼群与蚁群的混合算法去搭建的故障识别模型作比较。实验结果表明,该实验方法的有效性和可行性;

ACO算法 第2篇

Because of different social and cultural traditions, and different historical and geographic environment, people in different cultures have different ways of thinking, as a result of which, their aesthetic judgments are different.People in every culture have their own presupposition of what is beautiful and what is not, so what is regarded as beautiful in one culture may mean nothing in the other culture.The readers'expectation of beauty in translation must be met.If a translator does not take this into consideration, the beauty of the original work will be lost or will contradict to the expectation of readers;his translation will be surely rejected by the targeted readers.Since China and the Western world have different traditions, and Chinese people and people in the English world have different aesthetic values, which will accordingly influence people's way of writing and way of appreciating it.This paper will make a contrastive study of aesthetic values between China and the West and the effects caused by such differences on the Chinese to English translation and the English to Chinese one.

First of all, the philosophy of a culture has much influence on people's aesthetic judgments.Traditional Chinese philosophy such as Confucianism, Taoism, and Buddhism, has shaped Chinese people's way of thinking.Chinese people believe that the nature and man are two aspects of one thing, and that the subject is fused with the object.They like to express their feelings through the description of objects or imageries, and everything is given some subjective meanings in the Chinese writing (Jia2000:231) .As a result, the language used is usually very implicit, profound, picturesque and poetic.For example:

曲曲折折的荷塘上面, 弥望的是田田的叶子。叶子出水很高, 像亭亭的舞女的裙。层层的叶子中间, 零星地点缀着些白花, 有婀娜地开着的, 有羞涩地打着花骨朵儿的;正如一粒粒的明珠, 又如碧天里的星星, 又如刚出浴的美人。微风过处, 送来缕缕清香, 仿佛远处高楼上渺茫的歌声似的。这时候叶子与花也有一丝的颤动, 像闪电般, 霎时传过荷塘的那边去了。叶子本是肩并肩密密地挨着, 这便宛然有了一道凝碧的波痕。叶子底下是脉脉的流水, 遮住了, 不能见一些颜色;而叶子却更见风致了。 (Ed.Qiao, 2002:72)

In the Western world, on the other hand, people's way of thinking is shaped by the philosophy of ancient Greece, which seeks to explore the ultimate truth of the world.According to Plato, the physical world is the imitation of the ideal world, and the poetry is the imitation of the ideal world.There is a tradition of aesthetics of imitation;people in the English world stress"the imitation of the nature" (Jia, 2000:232) .According to Aristotle art can be more real than the physical world, and there is artistic truth (Ibid.) .As a result, concise, concrete, and coherent are stressed in writing in English.And the description is usually expected to conform to the objective world, and the images are usually explicit and simple.Implicit and overlapping language must be avoided.The following example illustrates the points.

Then came the twilight colors of sea and heaven, the wine pink width of water merging into lawns of aquamarines, the sky a tender palette of pink and blue. (Ed.Qiao, 2002:396)

Therefore, in order to make a translated text acceptable, enjoyable to the target readers, the difference mentioned above cannot be neglected in translation.When Duan Liancheng translated the Chinese example above literally and presented it to an American journalist, the latter made such a comment as"full of hyperbole", "not understandable", and"laughable" (Wang, 1998:6) .This is a good example to show that the imagery, implicit description that is considered beautiful by Chinese people does not meet American people's expectation of aesthetic values.Based on the foregoing analysis, the paper further differentiates two aspects for contrastive study.

Firstly, in Chinese writing, especially in artistic writing, people like to create a lot of images;they intend to express their feelings, their ideas through them.The objects are the emblems of men's feelings, ideas;and feelings and settings are often blended.Their writing reflects the beauty of imagery, while people in the West usually stand outside objective world to convey the beauty of it.People usually describe things as they are, their description is more rational, simple, natural, and without much implication.It is hard to overcome the difficulty caused by such difference.In order to satisfy the readers'expectation of beauty, adaptation is sometimes inevitable.The following example proves this.

The rain now becomes a torrent, flung capriciously by a rising wind.Together they batter the trees and level the grasses.Water streams off roofs and out of rain spouts.It pounds against the window in such a steady wash that I am sightless. (Ed.Qiao, 2002:272)

Version 1:这时, 雨变成了急流, 被风吹得到处乱飘。风和雨重重地打在地上, 打平了草地。雨水从屋顶上流下来, 漫出了排水管, 不停地敲打着窗户, 使我什么也看不清楚。眼前只有水。

Version 2:这时, 雨下得简直是倾盆如注, 狂风吹得雨水飘摇不定。风雨交加, 恣意地抽打树木, 夷平草地。雨水从屋顶奔流而下, 漫出了排水管, 不停地泼洒在窗户上, 使我的视野一片朦胧。眼前只有滔滔不绝的雨水。 (Ed.Qiao, 2002:273)

The description of the scene of the original work is more objective and simple, while in the second version, adjectives are added to modify each of the objects to add subjective meaning to make the images more vivid, which agrees with Chinese way of description.If we translate the passage word for word like the first version, in the eyes of Chinese readers, there is nothing beautiful.

Secondly, language is usually very implicit, vague, and full of hyperbole in Chinese writing;beautiful words or phrases are preferred by Chinese writers from ancient time till now, the meaning is usually not very clear or comprehensible.Chinese people value this implicit and unspeakable beauty.While in the West, the writing is much more explicit.People also apply hyperbole or other types of figures of speech;at the same time, they try to make their writing more direct and clear, and there is more deductive reasoning.All these are the factors that we must bear in mind in translation.For example:

大自然的鬼斧神工, 造就了长江三峡绝妙的奇景。气势雄伟的瞿塘峡, 逶迤曲折的巫峡, 礁石纵横的西陵峡, 无不风姿绰约, 光彩照人;深藏其间的小三峡, 更是曲水通幽, 楚楚动人, 山翠, 水清, 峰奇, 瀑飞, 倾倒了天下多少游人。 (Ed.Qiao, 2002:190)

The translated version:The Yangtze River boasts of the fascinating Three Gorges created by Mother Nature in all her glory The Three Gorges consist of grandiose Qutang Gorge, meandering Wu Gorge and heavily-shoaled Xiling Gorge, all saturated in splendid colors.Tucked away among these are three little gorges All of them form a veritable wonderland of clear water with plunging waterfalls and velvety hillsides, often rising to fantastic peaks-beckoning travelers from afar. (Ed.Qiao, 2002:191)

Both the original and the English convey the beauty of what people have expected.The Chinese text presents a spectacular scene to us through subjective description, while English text is more concise and smooth.

Besides philosophy, the aesthetic values are also affected by the moral standard of a culture.Confucianism has dominated Chinese moral standard for thousands of years.Traditionally talking about such things as money, sex is regarded as intolerable.Man and women used to express their love in an indirect way;they had no freedom to choose their love;pregnancy before marriage used to be viewed as a shame not only to oneself, but also to the whole family.Talking about such things in writing openly would offend the general readers.While in the West, the pursuit of freedom, happiness, and love is considered as an unalienable human right;pregnancy before marriage is understandable.There is nothing ugly or disgraceful.For this reason, Chinese translators in the past used to adapt the original texts to please readers.Look at the following sentences quoted from Romeo and Juliet.

He made you a highway to my bed

But I, a maid, die maiden-widowed

Version:他要你 (软梯) 做牵引相思的桥梁, 可是我却要做一个独受空房的怨女而死去 (朱生豪译)

In ancient China a man and a woman were not free to choose their spouse, sleeping together before marriage was considered erotic and could not tolerated by anyone.In order not to offend readers, and not to bring shame to the hero and the heroine, Zhu Shenghao presented the above-mentioned version.

Another difference in aesthetic values between China and the West is people's aesthetic expectation of literary norms (Wang, 1998:6) .If people of a certain culture are accustomed to certain forms of writing, other forms of writing will not be acceptable.For example, novels in the West have undergone a long history of development and have gained various forms, styles, and schools.While at the turn of the 20th century, novels had only very limited forms and styles in China.At that time each chapter of Chinese novels was expected to have a couplet title giving the gist of the content, it was started with such words as"the story goes like";and ended with such a sentence as"If you want to know what happens later, please read the next chapter".The first person point of view was seldom applied.For this reason, when Chen Duxiu, Su Manshu and some other Chinese translators translated literary works from some European language into Chinese, they added a couplet title to each chapter of the novels and they began each chapter with words like"the story goes"and ended each chapter with words like"if you like to know what happens later, please read the next chapter'.When Yan Fu translated Evolution and Ethics and Other Essays and Lin Shu translated La Dame aux Camelias, they changed the first person point of view to the third person of view.In the past, lengthy description of the character, and the long psycho-analysis of a figure were seldom seen in Chinese literature, while It was very a common practice in English literature.In order to please Chinese readers, when Fu Donghua translated Gone with the Wind, he deleted many long descriptions."I mercilessly deleted a whole chapter of lengthy description and psycho-analysis that is considered as insignificant in the development of the plot and tha makes readers bored" (Wang, 1998:6) .There is another example;in David Copperfield, there is a long description of 127words, Lin Shu abridged them into one short sentence.

The last aspect this thesis will focus on is the beauty of the arrangement of words, or the beauty of the form.According to Xu Yuanchong, we must attach importance to three aspects of beauty in translation:beauty of form, beauty of sound, and beauty o meaning (1998:401) .The following will discuss the first aspec mentioned by Professor Xu.

In successful Chinese writing, sentences are usually short grammatically loose;prepositions and conjunctions are seldom applied;four characters phrases or sentences are preferred;repetition of certain words or phrases is appreciated;and such figures of speech as parallel, antithesis are frequently used (Lian, 199364-65) .While in English writing, sentences are much longer, the grammatical relations are more complicated;prepositions and conjunctions are frequently used.The constituents in sentences paragraphs are grammatically closely connected.Repetition is not appreciated;people usually apply such means as substitution reference, and ellipsis to avoid repetition.Parallel and antithesis are not preferred.The following two examples serve to illustrate the point.

(1) .夜幕降临, 乐山上空礼炮轰隆, 彩灯光芒四溢。河面各只彩船上无数五颜六色的霓虹灯、日光灯齐放光华, 闪烁迷离, 和水中倒影连成一片。漂灯点缀河面, 似万点繁星坠落人间。在奇光异彩交相辉映中, 众多的龙舟如银河流星, 顺河而下, 千姿百态, 令人叫绝。在探照灯的扫射下, 河面激起变幻莫测的五彩霞光。一时间, 呈现出一副光的天地、色的世界。 (Ed.Qiao, 2002:182)

The translated version:As night falls, salutes from the city cannon thunder across the sky, a signal for countless bright lights to flash alive on all the dragon boats up and down the river Those myriad-lighted boats become a riot of color which blends most subtly with colors from the river's reflected surface.As the dragon boats row slowly and silently downstream, search lights along the shore highlight the procession of sparkling beauty.For a moment, before all fades to night, the spectators both afloat and ashore share this beautiful picture of universal brightness and magical color. (Ed.Qiao, 2002:183)

(2) She felt like a chess player who, by the clever handling of his pieces, sees the game taking the course intended;and her eyes were bright and tender with a smile as they glanced up into his;and her lips looked hungry for the kiss which they invited (Ed.Qiao, 2002:408)

The translated version:她觉得自己就像一个棋手, 巧妙地调动了每一颗棋子, 看着棋局按自己预定的要求发展。此刻她的眼睛又明亮又温柔, 含着笑意抬头望着他的眼睛, 饥渴的双唇迎候着他的吻。 (Ed.Qiao, 2002:409)

The analysis above is just some aspects of difference of aesthetic values between Chinese culture and Western culture and the application of the analysis in translation, which is something we should never neglect in translation.Meanwhile, the difference should not be overemphasized, the author of the paper here calls special attention to the following three points.First, people in different cultures have different aesthetic expectation, at the same time, they may share something similar, what is beautiful in one culture may be also enjoyable in the other culture.Second, aesthetic values is not absolute and unchangeable, what is considered beautiful in one culture at one time may not acceptable at the other time, what is intolerable at one time may be popular at the other time.Third, with the development of cross-culture communication, people's cultural presupposition, patterns of behaviors, and interpretation of events will be surely influenced by other cultures and will inevitably undergo some changes.As Nida points out:

But the revived Hebrew language of today is just no impediment to scientific thinking on the part of Israelis, and Greek was no abstract to the inductive reasoning of the Pre-Socratic philosophers nor did it later prevent Neo-Platonists from using Greek to promote their philosophic"flights of fantasy". (1993:106)

As a result, aesthetic values cherished by people of one culture will be more and more acceptable and appreciable in the other culture.Though there is always some difference between two types of writing, Chinese implicit imagery writing will surely be enjoyed in English world, and English explicit, abstract, deductive writing will be more and more acceptable in China.

参考文献

[1]Newmark, Peter.Approaches to Translation[M].Oxford:Pergamon Press, 1981.

[2]Nida E A.Language, Culture, and Translating[M].Shanghai:Shanghai Foreign Language Education Press, 1993:100-108.

[3]贾文波.从汉英景物描写看民族审美差异——从一段旅游景介英译所想到的[M].上海:上海外语教育出版社, 2000:230-241。

[4]连淑能.英汉对比研究[M].北京:高等教育出版社, 1993.

[5]刘宓庆.当代翻译理论[M].北京:中国对外翻译出版公司, 1999.

[6]乔萍, 瞿淑蓉, 宋洪玮.散文佳作108篇[M].南京:译林出版社, 2002.

[7]许钧.翻译思考录[M].武汉:湖北教育出版社, 2002.

ACO算法 第3篇

天然二氧化碳提纯是对天然CO2进行化工生产的传统工艺, 其主要原料主要有石灰窑气与石油喷井气体等, 原料气中主要含有游离水、烃类等杂质, 根据采气的压力和采气工艺决定, 从井口采出的二氧化碳是以液态形式存在, 其中所含的水和油脂也是以液态形式存在, 其余的杂质溶于二氧化碳液中。整个工艺过程通过汽化、吸附脱水、除尘、脱硫、干燥、变压吸附、提纯等单元操作对原气进行提纯净化, 生产出食品级和工业级二氧化碳产品, 其中提纯塔提纯是CO2生产过程中的最后一个环节, 也是最重要的一个环节, 主要是通过蒸馏原理进一步除去氮氧化合物及其它气体杂质, 以提高成品CO2的浓度。由于成品CO2在存放及运输过程中会有一部分氮氧化物自然分离, 按厂家目前的产品质量要求, 提纯塔出口CO2的浓度只需要达到99.5%以上即可。而在CO2提纯塔中影响成品浓度的主要有CO2的进塔浓度 (Input Chroma) 、进塔温度 (Input Temperature) 、塔顶压力 (Top Stress) 、加热温度 (Top Temperature) 、塔釜压力 (Bottom Stress) 及塔釜温度 (Inside Temperature) 7个因素, 各因素之间也存在复杂关系。

目前对天然CO2提纯控制方法中, 郑启富等人[1]提出建立成品CO2纯度与其影响因素之间的多元线性回归模型, 得出提纯塔CO2纯度的估算公式, 对工艺控制作用不够明显, 拟合精度较低, 刘欢培等人[2]建立了出口CO2纯度与其影响

因素之间的对数回归模型, 但回归模型及相关系数显著性不足, 影响模型的拟合程度, 对出口CO2纯度的控制效果不明显。周明华[3]提出BP神经网络方法及郑启富等人[4]又提出径向基神经网络与偏最小二乘回归相结合的方法 (RBFN-PLSR方法) , 两种方法所得结果均不够理想, 且存在训练速度慢, 易陷于局部极小和全局搜索能力弱等传统神经网络共同的缺点。

蚁群算法是一种新型的模拟进化算法, 它是由M.Dorigo于1992年在他的博士论文中引入, 其灵感来源于蚂蚁在寻找食物过程中发现路径的行为。由于该方法具有正反馈、分布式计算和富于建设性的贪婪启发式搜索特点, 所以应用越来越广泛, 从最初的TSP问题渗透到其他领域。洪炳熔等人[5]将蚁群算法和神经网络结合起来, 做了线性系统辨识问题及倒立摆控制问题的试验, 并得到了较好的仿真结果。本文利用蚁群算法优化神经网络的权值, 得到蚁群神经网络 (ACO-BP神经网络) 可兼有神经网络的广泛映射能力和蚁群算法的快速、全局收敛以及启发式学习等特点, 在某种程度上避免了神经网络收敛速度慢, 易陷入局部极小点的问题。

1、ACO—BP算法的原理与算法

1.1 ACO—BP算法的基本思路

假设网络中有m个参数, 包括所有权值和阀值。首先对这些参数进行排序, 将神经网络参数设置为N个随机的非零值, 从而形成集合IP, 然后定义一定数量的蚂蚁, 从蚁巢出发去寻找食物;每只蚂蚁都从第一个集合出发, 根据集合中每个元素的信息状态按照一定的概率从每个集合中选择一个元素, 当蚂蚁在所有集合中完成元素的选择后, 它就到达了食物源, 并按原路径返回蚁巢, 同时按下式 (1) 更新集合中所选元素的信息素, 这一过程反复进行, 当全部蚂蚁收敛到同一路径时, 也就意味着找到了网络参数的最优解。

表示t时刻集合IPi中元素j上的信息量, ρ表示信息素的维持因子, 表示所有蚂蚁在集合IPi中元素j上所释放的信息素的总和, 其表达式为

其中, ek表示将第k个蚂蚁选择的一组权值作为神经网络权值的输出误差, 式中o和oq表示神经网络的实际输出和期望输出。

1.2 ACO—BP算法的基本步骤

(1) 初始化:令时间t和迭代次数NC为零, 设置最大迭代次数NCmax, 随机产生一组权阈值集合IPi (1≤i≤m) , 令每个集合中的每个元素在初始时刻的信息素为0, 将所有蚂蚁置于蚁巢。

(2) 启动所有蚂蚁, 每只蚂蚁从第1个集合开始, 按照下述规则依次在每个集合中选择一个元素。路径选择规则:对于集合IPi任意一只蚂蚁k (k=1, 2, , s) , 根据 (2) 式计算的概率随机地选择它的第j个元素, 设表示在集合IPi中t时刻蚂蚁k由元素i转移到元素j的概率。

用以记录蚂蚁k已经选择的元素, 称为禁忌表, 它随着进化过程动态调整, allowed是不在禁忌表中的集合。α为信息启发式因子, 表示路径ij上残留信息在概率计算中的重要程度;β为期望启发因子, 反应了能见度ηij的相对重要程度;

(3) 重复步骤 (2) 直到所有蚂蚁都到达食物源。

(4) 依据各蚂蚁选择的权值计算BP神经网络的实际输出和误差, 记录当前最优解。所有蚂蚁按照原路径返回蚁巢, 设该过程经历的时间为n个时间单位, 则所选择的元素的信息素按 (1) 式做相应调节。

(5) 置t:=t+m, NC:=NC+1, 重复上述步骤, 直到进化趋势不明显或迭代次数NC≥NCmax, 则迭代结束, 输出最优解。否则转步骤 (2) 。

2、模型建立

2.1 数据处理及样本的确立

本次模型数据均采自浙江省某大型化工企业生产一线, 在半负荷状态下26天每隔一小时采集一次, 共采集618组数据, 限于篇幅, 数据不列, 使用拉依达准则测定不存在异常数据。选取568组数据作为训练样本, 另外50组数据作为检验数据。在软件运行中我们对数据进行了归一化处理。

2.2 神经网络结构设计

1989年Robert Hechi Nielson证明了一个三层的BP网络可以完成任意的N维到M维的映射[6], 所以一般采用一个隐含层的神经网络模型即可, 因此在神经网络控制器中选取三层网络结构。神经网络的输入层神经元数由影响因素确定, 输入节点个数选为7, 将输出层的节点数设为1, 取值范围为[-1.5, 1.5], 作为衡量输出CO2浓度的量化值。对于三层前向网络隐含层节点数, 本文采用经验公式 (m为隐含层节点数, n为输入层节点数, l为输出节点数, 为1~10之间的常数) , 根据实际情况, 确定隐节点个数为12个。即建立了网络结构为[7, 12, 1]。结合本文实际, 输入层、隐层分别采用对数型函数和线性函数。在神经网络的结构已经确定的情况下, 采用ACO-BP方法训练神经网络, 目标函数设定为当训练样本仿真输出和实际输出的误差平方和 (MSE) 下降到10-6时网络达到稳定状态, 训练结束, 由此运用该网络模型可预测提纯塔出口C O2浓度。

2.3 ACO—BP算法中参数的选定

为了对该网络模型进行验证, 笔者做了大量的计算机模拟实验, 选择蚂蚁群规模M=500, 权值区间为[-1.5, 1.5], 最大迭代次数设置为1000次, 网络的具体参数设置情况如表1所示。

表1 ACO-BP算法参数设置表

3、模型仿真与结果分析

利用Matl ab软件对数据进行运算, 通过多次实验比较, 并与传统BP预测方法进行比较 (如图1) 。

从比较图可以看出, 采用ACO-BP算法比使用传统BP预测拟合程度要高。采用剩余平方和检验拟合度可得ACO-BP预测SSR=5.4523, 而BP预测SSR=7.3158, ACO-BP预测误差系数r=1.547, 而BP预测误差系数r=1.814, 从这两方面均可知ACO-BP算法比使用传统BP预测拟合要好, 从而可以得出下列结论:本文给出的基于蚁群优化算法的三层前向反馈神经网络 (BP神经网络) 的训练模型, 与标准BP算法相比较, A CO-BP算法具有较快的收敛速度, 能够达到较小的均方误差值, 同时说明模型具有很强的拟合逼近能力及很好的预测效果。

4、结论

提纯塔中出口CO2浓度与其塔釜中各影响因素之间存在复杂的非线性关系, 本文利用ACO算法的快速、全局收敛以及启发式学习等特点与BP算法广泛映射能力, 为提纯塔出口CO2浓度提出ACO—BP建模方法, 对CO2提纯塔控制模型有较好的拟合效果, 对现实生产具有指导意义, 实现生产过程的优化控制。

摘要：利用ACO算法的快速、全局收敛以及启发式学习等特点与BP算法广泛映射能力, 为提纯塔出口CO2浓度提出ACO—BP建模方法, 对CO2提纯塔控制模型有较好的拟合效果, 对现实生产具有指导意义, 实现生产过程的优化控制。

关键词：蚁群算法,提纯搭,二氧化碳,BP神经网络,拟合

参考文献

[1]郑启富, 潘向军, 罗晟.C O2提纯塔建模研究[J].天津化工.2003, 17 (5) :2125.

[2]刘欢培, 张有正, 杨锡宝等.食品级二氧化碳提纯塔优化控制模型[J].浙江工业大学学报.2003, 31 (3) :332335.

[3]周明华, 汪国昭.神经网络在二氧化碳提纯塔建模中的应用[J].浙江大学学报 (工学版) .2006, 40 (2) :352:355.

[4]郑启富等.基于RBFN-PLSR方法的CO2提纯塔模型[J].化学工程.2005, 34 (4) :36:38.

[5]洪炳熔等.基于蚁群算法的多层前馈神经网络[J].哈尔滨工业大学学报.2003, 35 (7) :823:825.