外文翻譯---供應(yīng)鏈管理環(huán)境下的庫(kù)存優(yōu)化

1、<p><b>　　外文翻譯</b></p><p>　　題目：供應(yīng)鏈管理環(huán)境下的庫(kù)存優(yōu)化</p><p>　　摘要：傳統(tǒng)供應(yīng)鏈模式下的庫(kù)存優(yōu)化由于缺乏必要的信息，在模型的求解過(guò)程中難以得到符合實(shí)際要求的最優(yōu)解。本文分析了傳統(tǒng)企業(yè)庫(kù)存優(yōu)化與供應(yīng)鏈管理環(huán)境下庫(kù)存優(yōu)化的運(yùn)作機(jī)理，提出在供應(yīng)鏈管理環(huán)境下可以借助多層BP神經(jīng)網(wǎng)絡(luò)改進(jìn)傳統(tǒng)庫(kù)存模型，以得到更為滿意地最優(yōu)

2、庫(kù)存策略。并依據(jù)某一鋼材現(xiàn)貨公司的庫(kù)存情況給出具體的應(yīng)用。</p><p>　　關(guān)鍵詞：供應(yīng)鏈 供應(yīng)鏈管理 庫(kù)存 BP神經(jīng)網(wǎng)絡(luò) 優(yōu)化</p><p><b>　　1.引言</b></p><p>　　供應(yīng)鏈管理（簡(jiǎn)稱SCM）是當(dāng)今的一個(gè)熱門話題。這個(gè)詞來(lái)自關(guān)于作為一個(gè)特定的公司是如何組織聯(lián)系在一起的一幅圖片。供應(yīng)鏈管理的想法是采用整體的方法來(lái)管

3、理整個(gè)信息流，材料和來(lái)自于原材料供應(yīng)商的服務(wù)通過(guò)工廠和倉(cāng)庫(kù)直到最終的客戶。成功的供應(yīng)鏈管理需要有一個(gè)一體化的系列活動(dòng)納入一個(gè)緊密無(wú)間的過(guò)程。但是，在供應(yīng)鏈的每一個(gè)環(huán)節(jié)必然有一些延誤和一些不確定性，因此必須保持必要的庫(kù)存。相反，對(duì)企業(yè)來(lái)說(shuō)存貨實(shí)際上是一種浪費(fèi)。國(guó)內(nèi)外專家在庫(kù)存優(yōu)化領(lǐng)域已取得了很大的研究，做了許多庫(kù)存優(yōu)化的模型。所有這些模式在供應(yīng)鏈管理的思想應(yīng)運(yùn)而生之前已經(jīng)取得了，但這些模型沒(méi)有考慮上游和下游企業(yè)。這些作為稀缺的信息優(yōu)化模型

4、僅僅利用概率模型來(lái)適應(yīng)信息統(tǒng)計(jì)基礎(chǔ)上需求的變化。通常情況下，通過(guò)這種方式制作的模型因?yàn)樘^(guò)復(fù)雜而很難操作。另一方面，影響存貨清單的各因素之間的關(guān)系是非線性的，因此很難作出一個(gè)定量和明確的數(shù)學(xué)關(guān)系，而且這些最佳的成果也不能滿足實(shí)際應(yīng)用。</p><p>　　人工神經(jīng)網(wǎng)絡(luò)本身的自我學(xué)習(xí)和多映射的能力，可以探索復(fù)雜系統(tǒng)，使復(fù)雜的模型簡(jiǎn)單化。在人工神經(jīng)網(wǎng)絡(luò)里，隱藏在網(wǎng)絡(luò)中的信息所作的聯(lián)系的神經(jīng)元，它可以處理多種定量關(guān)系。

5、即神經(jīng)網(wǎng)絡(luò)是一個(gè)大規(guī)模并行計(jì)算模型，它的特點(diǎn)： </p><p>　　很大程度的魯棒性和容錯(cuò)性；

6、 </p><p>　　隨時(shí)準(zhǔn)備處理與一般非線性系統(tǒng)的相關(guān)問(wèn)題；</p><p><b>　　生物物理影響。</b></p><p>　　因此，對(duì)于非線性問(wèn)題，人工神經(jīng)網(wǎng)絡(luò)是一個(gè)很好的分析工具。本文將提出在多層次的BP神經(jīng)網(wǎng)絡(luò)的幫助下，來(lái)改進(jìn)傳統(tǒng)的庫(kù)存模型以獲得更令人滿意的優(yōu)化庫(kù)存。</p><p>　　2.傳統(tǒng)庫(kù)

7、存優(yōu)化模型的局限性</p><p>　　上游和下游企業(yè)形成之前的戰(zhàn)略聯(lián)盟關(guān)系，只有一個(gè)單一的物質(zhì)流。運(yùn)行機(jī)制如下圖所示：</p><p>　　基于傳統(tǒng)供應(yīng)鏈的運(yùn)行機(jī)制(如圖所示) ，由于缺乏必要的信息，庫(kù)存決策優(yōu)化模型必須利用概率模型來(lái)適應(yīng)信息統(tǒng)計(jì)基礎(chǔ)上需求的變化?，F(xiàn)在，我們給一個(gè)簡(jiǎn)單的單周期隨機(jī)庫(kù)存模型：</p><p><b>　　在這個(gè)模型中：<

8、;/b></p><p>　　E[T(y)]：價(jià)值期望的總費(fèi)用清單； c：每種產(chǎn)品制造（或購(gòu)買）的費(fèi)用；h：每個(gè)產(chǎn)品的庫(kù)存成本；p：缺少每個(gè)產(chǎn)品的懲罰成本；x:開(kāi)放的股票；y：該股在開(kāi)放時(shí)所得；ξ：在這個(gè)時(shí)期，它為隨機(jī)變量；φ（ξ）:概率密度函數(shù)。</p><p>　　為了盡量減少價(jià)值期望的總費(fèi)用清單的價(jià)值，即：使價(jià)值期望的總費(fèi)用清單最小，必須使。通過(guò)推導(dǎo)制定的方法得到參數(shù)的論點(diǎn)，我

9、們將得到：；如果提供每種產(chǎn)品制造（或購(gòu)買）的費(fèi)用，每個(gè)產(chǎn)品的庫(kù)存成本，缺少每個(gè)產(chǎn)品的懲罰成本的價(jià)值，我們能獲得該股在開(kāi)放時(shí)所得的最佳的價(jià)值股票，還可以得到在這個(gè)時(shí)代的最佳的庫(kù)存策略。</p><p>　　正如上面提到的，這種傳統(tǒng)模式下取得的資料不足，涉及到許多相關(guān)的應(yīng)用范圍，所以這是必不可少的前提假設(shè)，因此這種模式是難以符合實(shí)際應(yīng)用的。現(xiàn)在的主要問(wèn)題集中在隨機(jī)變量的概率密度函數(shù)中。從上述分析我們知道影響隨機(jī)變量的

10、因素是多變量非線性關(guān)系；如：產(chǎn)品的價(jià)格，銷售季節(jié)的變化，內(nèi)部收益率的總和。當(dāng)然，對(duì)于一個(gè)特定的企業(yè)、影響因素可能是可變的。因此，在這個(gè)時(shí)期的隨機(jī)變量可能不符合一個(gè)確定的概率分布，以及以這種方式獲得庫(kù)存的最優(yōu)戰(zhàn)略可能不符合現(xiàn)實(shí)的要求。</p><p>　　3.基于機(jī)械供應(yīng)鏈管理上的模型改進(jìn)</p><p>　　直接和深遠(yuǎn)影響到企業(yè)的供應(yīng)鏈變化的思考的決策模式: 改變傳統(tǒng)模式，阻止縱向思考模式

11、進(jìn)入橫向，縱向的思考模式打開(kāi)。隨著IT和物流技術(shù)的發(fā)展，基于內(nèi)聯(lián)網(wǎng)\聯(lián)網(wǎng)，互聯(lián)網(wǎng)和電子數(shù)據(jù)交換技術(shù)，企業(yè)可能有能力實(shí)現(xiàn)翻譯。在供應(yīng)鏈管理的基礎(chǔ)上機(jī)械業(yè)務(wù)的企業(yè)如下所示：</p><p>　　根據(jù)上圖中，屬于一個(gè)具體供應(yīng)鏈的企業(yè)可以分享一些重要的信息，這些信息在傳統(tǒng)供應(yīng)鏈下是每個(gè)企業(yè)的商業(yè)秘密。有了這一信息的企業(yè)可以提高庫(kù)存的預(yù)測(cè)精度，銷售等。</p><p>　　3.1多層BP神經(jīng)網(wǎng)絡(luò)&l

12、t;/p><p>　?。?）BP神經(jīng)網(wǎng)絡(luò)的摘要概括</p><p>　　人工神經(jīng)網(wǎng)絡(luò)，人工神經(jīng)網(wǎng)絡(luò)是一種信息處理模式啟發(fā)，通過(guò)密集的相互聯(lián)系，哺乳動(dòng)物大腦處理信息的平行結(jié)構(gòu)。換言之，人工神經(jīng)網(wǎng)絡(luò)的集合的數(shù)學(xué)模型，模擬的一些觀測(cè)特性的生物神經(jīng)系統(tǒng)，并利用類比的自適應(yīng)生物學(xué)習(xí)。神經(jīng)網(wǎng)絡(luò)模式的關(guān)鍵因素是新型結(jié)構(gòu)的信息處理系統(tǒng)。它是由大量的高度聯(lián)結(jié)處理單元，類似于捆綁在一起，以加權(quán)聯(lián)系，類似于突觸。這

13、一模式的優(yōu)勢(shì)尋找一個(gè)合適的預(yù)測(cè)模型庫(kù)存清單。</p><p>　　有眾多不同類型的人工神經(jīng)網(wǎng)絡(luò)和BP神經(jīng)網(wǎng)絡(luò)，這是進(jìn)行了反向誤差算法的訓(xùn)練。根據(jù)簡(jiǎn)單的結(jié)構(gòu)和大量的應(yīng)用，人工神經(jīng)網(wǎng)絡(luò)是目前最流行的神經(jīng)網(wǎng)絡(luò)。</p><p>　?。?）基本的多層BP神經(jīng)網(wǎng)絡(luò)</p><p>　　通常BP神經(jīng)網(wǎng)絡(luò)的層次是有組織的。層是由若干包含一個(gè)'激活功能'的相互關(guān)聯(lián)的

14、'節(jié)點(diǎn)'組成。模式通過(guò)“輸入層”提交給網(wǎng)絡(luò)，“輸入層”通過(guò)一個(gè)系統(tǒng)連接的加權(quán)對(duì)一個(gè)或更多的隱藏層進(jìn)行實(shí)際加工。隱藏層然后鏈接到一個(gè)‘輸出層’，在那里輸出所顯示的圖形如下：</p><p>　　BP神經(jīng)網(wǎng)絡(luò)的其他兩個(gè)要素是傳播fi，gi功能和神經(jīng)元之間的互連權(quán)重，即重：Wij，sij，和閾值的價(jià)值：θi，βi。這些元素之間的關(guān)系程度由方程式如下:</p><p>　　BP神經(jīng)

15、網(wǎng)絡(luò)包含一些通過(guò)輸入模式來(lái)修改權(quán)的連接的某種形式的'學(xué)習(xí)規(guī)則'。雖然有許多不同類型的學(xué)習(xí)規(guī)則，但三角洲規(guī)則是BP神經(jīng)網(wǎng)絡(luò)用的最常見(jiàn)的學(xué)習(xí)規(guī)則。在三角洲規(guī)則里， '學(xué)習(xí)'是出現(xiàn)在每個(gè)周期或'時(shí)代'的通過(guò)產(chǎn)出流動(dòng)激活以及重量調(diào)整誤差，向后傳播的一個(gè)監(jiān)督的過(guò)程。</p><p>　　3.2在隨機(jī)變量的基礎(chǔ)上制作BP神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)模型</p><p>

16、　　作為庫(kù)存優(yōu)化模型、關(guān)鍵元素是適應(yīng)隨機(jī)變量需求的變化。 同時(shí)，影響需求的因素是可變的，在這個(gè)意義上說(shuō)，他也是模型中最艱難的過(guò)程。另一方面，因?yàn)檫@些企業(yè)雙贏的關(guān)系，屬于一個(gè)具體供應(yīng)鏈的企業(yè)可以分享一些重要的信息。如:操作計(jì)劃,營(yíng)銷情報(bào)等信息。這些因素是非線性，為了使庫(kù)存優(yōu)化相當(dāng)精確，我們可以利用三重層BP神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)的變化著的預(yù)測(cè)模型。</p><p>　　制作BP神經(jīng)網(wǎng)絡(luò)庫(kù)存預(yù)測(cè)的關(guān)鍵部件是因素和量化的選擇。隨

17、即變量首先要求選擇的因素必須符合在隨機(jī)變量的基礎(chǔ)上制作BP神經(jīng)網(wǎng)絡(luò)預(yù)測(cè)模型，然后，我們也將考慮到定量的可行性?，F(xiàn)在我們根據(jù)鋼鐵企業(yè)實(shí)際的庫(kù)存條件給出一個(gè)具體的三重層BP神經(jīng)網(wǎng)絡(luò)模型來(lái)預(yù)測(cè)鋼板的需求。</p><p>　　鋼鐵公司也是供應(yīng)鏈的一個(gè)鏈接點(diǎn)，因此它可以從它的戰(zhàn)略合作者得到一些具體信息?，F(xiàn)在，我們選擇以往時(shí)代的需求：x1，這個(gè)時(shí)期的價(jià)格：x2，內(nèi)部收益率的鋼鐵行業(yè)的總數(shù)：x3，季節(jié)因素的變化：x4，四個(gè)因

18、素的輸入層；決策速度的需求v，替代的隨機(jī)變量，作為輸出層；同時(shí)，隱層神經(jīng)元的數(shù)目應(yīng)取決于我們所使用的優(yōu)化方法。該模型的結(jié)構(gòu)如下所示： </p><p>　　在一些采樣數(shù)據(jù)里，我們可以選擇一個(gè)合適的傳遞函數(shù)并且培訓(xùn)這種模式。在這個(gè)培訓(xùn)過(guò)程中，我們可以利用矩陣實(shí)驗(yàn)室提供的神經(jīng)網(wǎng)絡(luò)工具。一旦模型訓(xùn)練達(dá)到令人滿意的水平，我們可以利用它來(lái)預(yù)測(cè)本公司的庫(kù)存需求的變化。</p><p><b>

19、;　　4.結(jié)論</b></p><p>　　根據(jù)上述分析，很難適當(dāng)描述傳統(tǒng)做法下影響庫(kù)存需求的因素之間的關(guān)系。另外我們知道，神經(jīng)網(wǎng)絡(luò)善于解決一些沒(méi)有解決辦法或其中一個(gè)解決方案的算法太復(fù)雜而無(wú)法找到的問(wèn)題?？傮w來(lái)看，這個(gè)神經(jīng)網(wǎng)絡(luò)模型，特別是在目前用BP神經(jīng)網(wǎng)絡(luò)模型來(lái)預(yù)測(cè)庫(kù)存需求的變化的時(shí)候，是一種合適的方法。</p><p>　　INVENTORY OPTIMUM BASED O

20、N SUPPLY CHAIN MANAGEMENT</p><p>　　YUN Jun YAN Bing ZHAO Yuwei</p><p>　　College of Management of Wuhan University of Technology Hubei Wuhan</p><p>　　Abstract: Because th

21、e optimized inventory in traditional supply chain model has poor information, it becomes more difficult to obtain optimal solution complying with the practical requirements during finding, solutions to supply chain patte

22、rns. This article is intended to analyze the operational mechanism of optimized inventory in both traditional enterprises and supply chain management. Also，this article put forward to improve traditional inventory patter

23、ns with the aid of multiple-layer BP neu</p><p>　　Key Words: Supply Chain, SCM, Inventory, BP Neural Network, Optimized</p><p>　　1.INTRODUCTION</p><p>　　Supply chain Management (SCM

24、 for short) is a hot topic today. The term supply chain comes from a picture of how organizations are linked together as viewed from a particular company. The idea of SCM is to apply a total systems approach to managing

25、the entire flow of information, materials, and services from raw materials suppliers through factories and warehouses to the end customers . Successful SCM requires an integration of series activities into a seamless pro

26、cess. However, there must be som</p><p>　　Artificial neural network have the ability to learn by itself and multi-mapping, and it can explore complicate system escaping to make complicate models. In the arti

27、ficial neural network models, the information hides in the network made by linked-neuron, and it can deal with multiple quantitative relationships. Namely, the ANN is a massively parallel computational model, and it has

28、characterizes : </p><p>　　·Great degree of robustness and fault tolerance;</p><p>　　·Ready to deal with problems associated with general nonlinear systems;</p><p>　　·

29、Biophysical implications.</p><p>　　So ANN is a good analysis tool for nonlinear problem. This paper will put forward to improve traditional inventory models with the aid of multi- layer BP neural network so

30、as to acquire much more satisfactory optimum tactics of inventory.</p><p>　　2.THE LIMITATION TRADITIONAL INVENTORY OPTIMUM MODEL</p><p>　　Before the strategic alliance relationship among the ups

31、tream and downstream enterprises comes into being, there is only a single material flow. The operational mechanics is shown below:</p><p>　　Under the operational mechanies of traditional supply chain(as show

32、 in figurel)，making inventory optimurn models; because of the lack of the necessary information, have to utilize probabilistic models to fit the changes of requirements based on the information of statistics. Now we giv

33、e a simple single period random inventory model: </p><p>　　In this model:</p><p>　　E[T (y)] :The value of expectation of the total cost of inventory; c :The manufacture(or purchase)cost of per

34、product; h: The inventory cost of per product; p :The punishment cost for shorts of per product; x: The opening stock; y :The stock obtained at opening; ξ: The demand during this epoch, it is a random variable; φ(ξ):Th

35、e probability density function of ξ.</p><p>　　In order to minimize the value of E[T(y)],namely，this must have.Following the method of derivation formulation which obtains parameter argument, we will get ;If

36、 give the value of c , h , p ,we can get the optimum value of stock y' ;also we can get the optimum inventory tactics during this epoch.</p><p>　　As referred above, this traditional model is made under t

37、he insufficient information, so it is essential to lead many premise hypotheses, delimitate the application range, so this kind of model is difficult to accord with the application in the real-world. The main problem foc

38、us on the probability density functions of ξ .From the analysis above we know the factors which affect the random variable ξ are a multi-variable nonlinear relationship; such as: the price of product, the change of marke

39、tin</p><p>　　3.MODEL IMPROVEMENT BASED ON THE MECHANIC OF SUPPLY CHAIN MANAGEMENT</p><p>　　The direct and profound effect to the enterprise by the think of SCM is the change of decision mode:

40、 Change from the traditional, blocked longitudinal think mode into transversal, opening think mode. With the developing of IT and logistics technology, enterprise may have the ability to realize the translation based the

41、 Intranet\Extranet, Internet, and EDI technology .The operational mechanic of enterprise based on the SCM is shown below:</p><p>　　According to the figure above, enterprises to one specific SC may share som

42、e important information which is the business secret for enterprise under the information enterprise traditional SC. With this information enterprise can improve prediction for inventory, marketing, the precision of etc

43、.</p><p>　　3.1 MULTIPLE-LAYER BP NEURAL NETWORK</p><p>　　(1) Generalization of BPNN</p><p>　　Artificial Neural Network, ANN for short, is an information-processing paradigm inspired

44、 by the way the densely interconnected, parallel structure of the mammalian brain processes information. In other words, artificial neural networks are collections of mathematical models that emulate some of the obser

45、ved properties of biological nervous systems and draw on the analogies of adaptive biological learning. The key element of the ANN paradigm is the novel structure of the information processin</p><p>　　There

46、 are multitudes of different types of ANNs, and BP Neural Network is a multi-layers back propagation neural network, which is trained with the backpropagation of error algorithm. According to the simple structure and the

47、 considerable application, BPNN is the most popular ANN at the present.</p><p>　　(2)The basics of multi-layers BPNN</p><p>　　BPNN is typically organized in layers. Layers are made up of a number

48、 of interconnected 'nodes' which contain an 'activation function'. Patterns are presented to the network via the 'input layer', which communicates to one or more 'hidden layers' where th

49、e actual processing is done via a system of weighted 'connections'. The hidden layers then link to an 'output layer' where the answer is output as shown in the graphic below:</p><p>　　The o

50、ther two elements of BPNN are the propagation funtions of fi ,gi and the interconnection weights between the namely the weights :wij, sij，and the value of threshold: θi,βi .The relationships of these elements are deter

51、minated by the equations as follow:</p><p>　　BPNN contains form of 'learning rule' which modifies weights of the connections according to the input patterns. Although there are many different kinds

52、of learning rules, what BPNN uses the most often is the delta rule. With the delta rule, 'learning' is a supervised process that occurs with each cycle or 'epoch' through a forward activation flow of outp

53、uts, and the backwards error propagation of weight adjustments.</p><p>　　3.2 Making prediction BPNN model based on the random variable ξ</p><p>　　As to the inventory optimum model, the key eleme

54、nt is to fitting the change of the random variable of demand ξ, at the same time, the factors which affect the demand are variable, in this sense, it is also the most difficult process in the model. On the other hand ent

55、erprises which belonged to one specific SC can share some important information, due to the win-win relationships among these enterprises. The information such as: the operational plan, the marketing intelligence etc.

56、 These facto</p><p>　　The key component for making inventory prediction BPNN is the choice of influence factors and the quantification of them. First of all the criteria for the choice of factors must lie on

57、 the contribution rate for ξ，then we will also take account of the feasibility of quantification. Now we will give a specific triple-layers BPNN model based on the actual inventory condition of a steel corporation to pre

58、dict the change of the demand of the steel plate.</p><p>　　This steel corporation is also a link in a supply chain, so it can get some specific information from its strategic cooperators. Now we select the

59、demand of previous epoch: x1，the price of this epoch:x2，the internal rate of return of total steel vocation: x3，the factor of season change: x 4,making the four factors as input layer; making the demand velocity v,the su

60、bstitution of ξ，as output layer; At the same time, the number of the 'hidden layer' neuron should depend on the optimization method</p><p>　　With some sampled data, we can select a suitable transfer

61、 function and train this model. In the process of training, we can use the ANN tools provided by MATLAB. Once the model is trained to a satisfactory level, we can utilize it to predict the change of this corporation'

62、s inventory demand. To do this, we can get the next demand based on the current data.</p><p>　　4.CONCLUSION</p><p>　　According to the anaaysis above, it is difficult to describe adequately the r

63、elationship of the factors which affect the demand of inventory with conventional approaches. Also we know that the ANNs are good at solving problems that do not have an algorithmic solution or for which an algori