电气工程专业外文翻译-建模与模拟导论（含英文原文）.docx

建模与模拟导论Anu人玛丽亚纽约州立宾汉姆顿大学系统科学与工业工程系13902-6000宾厄姆顿，纽约，美国外文文献为PDF格式，下载后双击即可打开另存摘要这个入门教程讲的是仿真概述、modeling and analysis. Many critical questions are建模与分析。许多关键性的问题在answered in the paper. What is modeling? What is本文中得到了回答。什么是建模？什么是simulation? What is simulation modeling and analysis?模拟？仿真建模与分析是什么？What types of problems are suitable for simulation? How什么类型的问题适合模拟？怎么to select simulation software? What are the benefits and选择仿真软件？有什么好处，pitfalls in modeling and simulation? The intended建模与仿真中的陷阱？预期audience is those unfamiliar with the area of discrete受众是那些不熟悉离散区域event simulation as well as beginners looking for an事件模拟以及初学者寻找overview of the area. This includes anyone who is该领域概况。这包括involved in system design and modification - system参与系统设计和修改的系统analysts, management personnel, engineers, military分析师，管理人员，工程师，军事planners, economists, banking analysts, and computer规划师，经济学家，金融分析师，和计算机scientists. Familiarity with probability and statistics is科学家。用概率统计的熟悉assumed.假定。1什么是建模？建模是制作模型的过程；模型is a representation of the construction and working of是代表着一些建设的特征和some system of interest. A model is similar to but一些系统的兴趣。一个模型和现实系统相似，但simpler than the system it represents. One purpose of a比它所代表的系统要简单。一个目的之一是model is to enable the analyst to predict the effect of模型可以使分析师预测的一些变化产生的影响changes to the system. On the one hand, a model should对系统的影响。另一方面，模型be a close approximation to the real system and需接近incorporate most of its salient features. On the other将其真正的系统大部分的显著特征。另hand, it should not be so complex that it is impossible to一方面，它不应该是非常复杂的，以使它容易的understand and experiment with it. A good model is a理解和实验。一个好的模型是一个judicious tradeoff between realism and simplicity.现实和简单性之间明智的权衡。Simulation practitioners recommend increasing the模拟从业人员建议增加complexity of a model iteratively. An important issue in一个模型的复杂性迭代。一个重要的问题modeling is model validity. Model validation techniques模拟模型的有效性。模型验证技术include simulating the model under known input包括在输入已知的模型模拟conditions and comparing model output with system条件和比较模型输出与系统output.输出结果。一般来说，用于模拟研究模型is a mathematical model developed with the help of应是基于一个数学模型的simulation software. Mathematical model classifications仿真软件。数学模型的分类include deterministic (input and output variables are包括确定（输入和输出变量fixed values) or stochastic (at least one of the input oroutput variables is probabilistic); static (time is not taken固定值）或随机（在输入或输出变量中至少有一个概率变量）；静（时间不采取into account) or dynamic (time-varying interactions考虑）或动态（随时间变化的相互作用among variables are taken into account). Typically,在变量的考虑）。通常，simulation models are stochastic and dynamic.仿真模型是随机的、动态的。2模拟的是什么？模拟系统是一个模型的操作the system. The model can be reconfigured and系统。该模型可以被重新配置和experimented with; usually, this is impossible, too试验；通常，这是不可能的，太expensive or impractical to do in the system it represents.昂贵或不切实际的做系统的代表。The operation of the model can be studied, and hence,该模型的运行进行研究，因此，properties concerning the behavior of the actual system对实际系统的行为特性or its subsystem can be inferred. In its broadest sense,或其子系统可以进行推断。在其最广泛的意义上，simulation is a tool to evaluate the performance of a仿真是一种用工具来评估一个绩效system, existing or proposed, under different系统，现有的或提出的，在不同的configurations of interest and over long periods of real配置的兴趣和长期的time.时间里。使用模拟在现有系统altered or a new system built, to reduce the chances of改变或新制度的建立，为了减少failure to meet specifications, to eliminate unforeseen不符合规格的机会，排除不可预见的bottlenecks, to prevent under or over-utilization of瓶颈，为了利用resources, and to optimize system performance. For资源，并优化系统性能。对于instance, simulation can be used to answer questions例如，模拟可以用来回答问题like: What is the best design for a new如：一个新设计的最好的是什么telecommunications network? What are the associated电信网络？什么是相关的resource requirements? How will a telecommunication资源需求？如何将电信network perform when the traffic load increases by 50%?网络进行，当流量负载增加50%？How will a new routing algorithm affect its如何将一种新的路由算法的影响performance? Which network protocol optimizes性能？网络协议优化network performance? What will be the impact of a link网络性能？什么是链接的failure?失败影响？本教程的主题是离散事件simulation in which the central assumption is that the模拟的中心假设是system changes instantaneously in response to certain瞬时响应某些系统的变化discrete events. For instance, in an M/M/1 queue - a离散事件。例如，在一个M / M / 1排队-single server queuing process in which time between单台服务器的排队过程中，时间的关系到达arrivals and service time are exponential - an arriva和服务时间是指数的一个信号causes the system to change instantaneously. On the导致系统瞬时变化。在other hand, continuous simulators, like flight simulators另一方面，连续的模拟器，如飞行模拟器and weather simulators, attempt to quantify the changes和天气模拟器，试图量化的变化in a system continuously over time in response tocontrols. Discrete event simulation is less detailed在一个系统中连续两次响应。因此，在各种各样的情况下，离散事件仿真(coarser in its smallest time unit) than continuous（在最小的时间单位粗）比连续simulation but it is much simpler to implement, and模拟的不详细但它是非常简单的实现。hence, is used in a wide variety of situations.图1是一个模拟示意图。由系统显示的iterative nature of the process is indicated by the system迭代过程的性质是under study becoming the altered system which then在研究中成为改变的系统，然后becomes the system under study and the cycle repeats. In就所研究的系统和循环重复。在a simulation study, human decision making is required at一个模拟的研究，人类的决策是需要的all stages, namely, model development, experiment所有的阶段，即，模型的开发，实验design, output analysis, conclusion formulation, and设计，输出分析，结论和公式，making decisions to alter the system under study. The决定改变系统的研究。不需要人的干预the running of the simulations, which most simulation的模拟运行的only stage where human intervention is not required is唯一的平台，大多数模拟software packages perform efficiently. The important软件包的有效执行。重要point is that powerful simulation software is merely a的一点是，强大的仿真软件也会因一个hygiene factor - its absence can hurt a simulation study关健因素的缺失可以伤害的模拟研究，but its presence will not ensure success. Experienced但它的存在不会保证成功。有经验的problem formulators and simulation modelers and问题配方、仿真建模和analysts are indispensable for a successful simulation分析师对于一个成功的模拟是必不可少的study.研究。现实世界 模拟世界模拟模型结 论模拟分析模拟实验研究的系统改变系统图1：模拟示意图参与开发一个仿真的步骤model, designing a simulation experiment, and模型，仿真实验设计，和performing simulation analysis are:进行仿真分析：Step 1. Identify the problem.步骤1：确定问题。Step 2. Formulate the problem.步骤2：制定的问题。Step 3. Collect and process real system data.步骤3：真正的系统的数据收集和处理。Step 4. Formulate and develop a model.步骤4：制定和建立一个模型。步骤5：验证模型。Step 6. Document model for future use.步骤6：为未来使用文档模型。Step 7. Select appropriate experimental design.步骤7：选择适当的实验设计。Step 8. Establish experimental conditions for runs步骤8：建立运行实验条件Step 9. Perform simulation runs.步骤9：进行模拟运行。Step 10. Interpret and present results.步骤10：解释分析结果。Step 11. Recommend further course of action.步骤11：建议进一步的行动过程。虽然这是一个逻辑排序simulation study, many iterations at various sub-stages仿真研究，在不同的分阶段多次迭代may be required before the objectives of a simulation可以在模拟的目标要求study are achieved. Not all the steps may be possible研究了。并不是所有的步骤是可能的and/or required. On the other hand, additional steps may和、或要求。另一方面，额外的步骤have to be performed. The next three sections describe必须执行。接下来的三节描述these steps in detail.在详细介绍这些步骤。3如何开发一个模拟MODEL?模型？仿真模型包括以下组件：system entities, input variables, performance measures,系统实体，输入变量，性能的措施，and functional relationships. For instance in a simulation和功能的关系。比如在一个模拟model of an M/M/1 queue, the server and the queue are的M / M / 1排队模型，服务器和队列system entities, arrival rate and service rate are input系统实体，输入到达率和服务率variables, mean wait time and maximum queue length变量，平均等待时间和最大队列长度are performance measures, and time in system = wait性能的措施，和在系统时间=等待time + service time is an example of a functional时间+服务时间”是一个功能的一个例子relationship. Almost all simulation software packages关系。几乎所有的仿真软件包provide constructs to model each of the above为构建模型的上述各components. Modeling is arguably the most important组件。建模是最重要的part of a simulation study. Indeed, a simulation study is一个仿真研究。事实上，一个仿真研究as good as the simulation model. Simulation modeling作为仿真模型为好。仿真建模comprises the following steps:包括以下步骤：步骤1。确定问题。列举问题with an existing system. Produce requirements for a与现有的系统。生产要求proposed system.提出的系统。步骤2。制定的问题。选择范围of the system, the problem or a part thereof, to be该系统的问题，或其一部分，是studied. Define overall objective of the study and a few研究。定义和一些研究的总体目标specific issues to be addressed. Define performance要解决的具体问题。定义绩效measures - quantitative criteria on the basis of which措施的定量标准的基础上different system configurations will be compared and不同的系统配置进行比较，ranked. Identify, briefly at this stage, the configurations排名。确定，在这个阶段的配置简单，of interest and formulate hypotheses about system兴趣和制定假设系统performance. Decide the time frame of the study, i.e.,性能。决定时间的研究框架，即，will the model be used for a one-time decision (e.g.,将该模型用于一次性决策（例如，capital expenditure) or over a period of time on a regular资本支出）或在一段时间内有规律的basis (e.g., air traffic scheduling). Identify the end user基础（例如，空中交通调度）。确定最终用户of the simulation model, e.g., corporate management该仿真模型，例如，企业管理versus a production supervisor. Problems must be与生产主管。问题必须formulated as precisely as possible.制定尽可能精确的。步骤3。真正的系统的数据收集和处理。Collect data on system specifications (e.g., bandwidth for收集系统规范的数据（例如，带宽a communication network), input variables, as well asperformance of the existing system. Identify sources of通信网络），输入变量，作为现有系统性能。源的识别randomness in the system, i.e., the stochastic input在系统中的随机性，即，随机输入variables. Select an appropriate input probability变量。选择一个合适的输入概率distribution for each stochastic input variable and对于每一个随机输入变量的分布estimate corresponding parameter(s).估计相应的参数。软件包进行分布拟合selection include ExpertFit, BestFit, and add-ons in some选择包括expertfit，碧辉，和一些插件standard statistical packages. These aids combine标准的统计软件包。这些艾滋病结合goodness-of-fit tests, e.g., 2拟合优度检验，例如，test, Kolmogorov-Smirnov测试，Kolmogorov斯米尔诺夫test, and Anderson-Darling test, and parameter测试，和Anderson-Darling检验，参数estimation in a user friendly format.在一个用户友好的格式估计。标准分布，如泊松分布，指数，normal, hyperexponential, etc., are easy to model and正常，超指数，等等，都容易模型simulate. Although most simulation software packages模拟。虽然大多数的仿真软件包include many distributions as a standard feature, issues包括许多分布作为标准功能，问题relating to random number generators and generating关于随机数生成器和生成random variates from various distributions are pertinent随机变量的各种分布相关and should be looked into. Empirical distributions are并应考虑。经验分布used when standard distributions are not appropriate or使用时的标准分布不适当或do not fit the available system data. Triangular, uniform不适合可用的系统数据。三角形，均匀or normal distribution is used as a first guess when no或正态分布作为第一想当没有data are available. For a detailed treatment of probability数据是可用的。一个详细的治疗的概率distributions see Maria and Zhang (1997).分布看玛丽亚和张（1997）。第四步。制定和开发模型。发展schematics and network diagrams of the system原理图和系统网络图（如何做entities flow through the system实体流通过系统？）翻译这些。conceptual models to simulation software acceptable仿真软件可接受的概念模型form. Verify that the simulation model executes as形态。验证了仿验证了仿真模型的执行intended.预期。核查技术包括痕迹，变input parameters over their acceptable range and输入参数的可接受范围与checking the output, substituting constants for random检查输出，常、随机variables and manually checking results, and animation.检查结果变量与扳，和动画。步骤5。验证模型。比较模型performance under known conditions with the已知条件下的性能performance of the real system. Perform statistical实际系统的性能。进行统计inference tests and get the model examined by system推理测试得到的系统模型研究experts. Assess the confidence that the end user places专家。评估的信心，最终用户的地方on the model and address problems if any. For major在模型和解决问题，如果任何。主要simulation studies, experienced consultants advocate a仿真研究，经验丰富的顾问主张structured presentation of the model by the simulation结构化模型的模拟演示analyst(s) before an audience of management and system分析师在管理和系统的观众experts. This not only ensures that the model专家。这不仅保证了模型assumptions are correct, complete and consistent, but假设是正确的，完整的和一致的，但also enhances confidence in the model.也提高了模型的信心。步骤6。为未来使用文档模型。文件objectives, assumptions and input variables in detail.的目标，详细的假设和输入变量。4如何设计一个模拟EXPERIMENT?实验？仿真实验是一个测试或一系列的测试which meaningful changes are made to the inputvariables of a simulation model so that we may observe这有意义的改变是一个仿真模型的输入变量，我们可以观察and identify the reasons for changes in the performance并确定在性能变化的原因measures. The number of experiments in a simulation措施。在模拟实验次数study is greater than or equal to the number of questions研究大于或等于问题的数量being asked about the model (e.g., Is there a significant被问到的模型（例如，有一个显着的difference between the mean delay in communication在通信的平均延迟之间的差异networks A and B?, Which network has the least delay:网络A和B？其中，网络具有最小的延迟：A, B, or C? How will a new routing algorithm affect theA，B，或C？如何将一种新的路由算法的影响performance of network B?). Design of a simulation网络B的表现？）一个仿真设计。experiment involves answering the question: what data实验涉及到回答的问题：什么样的数据need to be obtained, in what form, and how much? The需要获得，以什么样的形式，又有多少？following steps illustrate the process of designing a以下步骤说明设计的过程simulation experiment.模拟实验。步骤7。选择适当的实验设计。Select a performance measure, a few input variables that选择一个性能指标，输入变量少，are likely to influence it, and the levels of each input可能影响它，和每一个输入的水平variable. When the number of possible configurations变量。当一些可能的配置(product of the number of input variables and the levels（输入变量数的产品和水平of each input variable) is large and the simulation model每个输入变量）是大型的仿真模型is complex, common second-order design classes是复杂的，常见的二阶设计类including central composite, Box-Behnken, and full-包括中央复合，Box-Behnken，全部factorial should be considered. Document the因应考虑、文件experimental design.实验设计。步骤8。建立运行实验条件。Address the question of obtaining accurate information地址获取准确信息的问题and the most information from each run. Determine if the从每一个运行的信息最。确定system is stationary (performance measure does not系统是静止的（性能指标不change over time) or non-stationary (performance随着时间的变化而变化）或非平稳（性能measure changes over time). Generally, in stationary随时间变化的测量）。一般来说，在固定systems, steady-state behavior of the response variable is系统，响应变量的稳态行为of interest. Ascertain whether a terminating or a non-感兴趣的。确定是否终止或非terminating simulation run is appropriate. Select the run终端仿真运行是合适的。选择运行length. Select appropriate starting conditions (e.g., empty长度。选择合适的启动条件（例如，and idle, five customers in queue at time 0). Select the空闲时间的0，五的客户在队列）。选择length of the warm-up period, if required. Decide the在预热期的长度，如果需要的话。决定number of independent runs - each run uses a different一些独立的运行，每次运行使用不同random number stream and the same starting conditions -随机数流和相同的初始条件by considering output data sample size. Sample size must考虑到输出数据的样本大小。样本必须be large enough (at least 3-5 runs for each configuration)要足够大（至少3-5运行每个配置）to provide the required confidence in the performance提供所需的性能的信心measure estimates. Alternately, use common random估计。另外，使用常见的随机numbers to compare alternative configurations by using a数字用一个比较另类的配置separate random number stream for each sampling每个独立的随机数流采样process in a configuration. Identify output data most在配置过程。确定输出数据likely to be correlated.可能是相关的。步骤9。进行模拟运行。执行运行according to steps 7-8 above.根据以上步骤7-8。5如何进行仿真ANALYSIS?分析？大多数的仿真软件包提供运行统计（平均，standard deviation, minimum value, maximum value) on标准偏差，最小值，最大值）在the performance measures, e.g., wait time (non-time性能的措施，例如，等待时间（非时间persistent statistic), inventory on hand (time persistent持续的统计），存货（时间持续statistic). Let the mean wait time in an M/M/1 queue统计）。让平均等待一个M / M / 1排队时间observed from n runs be从N运行观察n 2 1 W ., W , W . It is important toW1，W2，W3。这是很重要的understand that the mean wait time W is a random，平均等待时间是随机的variable and the objective of output analysis is to变量和输出分析的目的是estimate the true mean of W and to quantify its估计W的真值和量化其variability.变异性。尽管事实是没有数据collection errors in simulation, the underlying model is在仿真模型的基础是收集错误，fully known, and replications and configurations are user充分认识，并复制和配置的用户controlled, simulation results are difficult to interpret. An控制，仿真结果难以解释。一个observation may be due to system characteristics or just观察可能是由于系统的特性或是a random occurrence. Normally, statistical inference can一个随机事件。通常情况下，统计推断assess the significance of an observed phenomenon, but评估一个观察到的现象的意义，但most statistical inference techniques assume大多数的统计推断技术假设independent, identically distributed (iid) data. Most types独立同分布（IID）的数据。大多数类型的of simulation data are autocorrelated, and hence, do not仿真数据的自相关，因此，不satisfy this assumption. Analysis of simulation output满足这个假设。仿真输出分析data consists of the following steps.数据包括以下步骤。步骤10。解释分析结果。计算numerical estimates (e.g., mean, confidence intervals) of（例如，平均数值估计，置信区间）的the desired performance measure for each configuration对于每个配置所需的性能测量of interest. To obtain confidence intervals for the mean感兴趣的。获得均值的置信区间of autocorrelated data, the technique of batch means can的相关数据，对一批技术手段可以be used. In batch means, original contiguous data set使用。在批处理方式，原来的连续数据集from a run is replaced with a smaller data set containing从运行一个较小的数据集包含取代the means of contiguous batches of original observations.对原观测连续批次的手段。The assumption that batch means are independent may假设批手段是独立的可能not always be true; increasing total sample size and不一定是真的；增加总样本的大小和increasing the batch length may help.增加批长度可以帮助。测试系统性能研究。Construct graphical displays (e.g., pie charts, histograms)构建图形显示器（例如，饼图，直方图）of the output data. Document results and conclusions.输出数据的。文献的研究结果和结论。步骤11。建议进一步的行动过程。这may include further experiments to increase the precision可能包括进一步的实验来提高精度and reduce the bias of estimators, to perform sensitivity减少估计偏差，进行敏感性analyses, etc.分析，等等。6个例子机加工车间有两钻，一个矫直机，和one finishing operator. Figure 2 shows a schematic of the一个整理算子。图2显示了一个示意图machine shop. Two types of parts enter the machine shop.机加工车间。两种部分进入车间。矫直机 整理算子一号钻二号钻 图2：对车间示意图1型零件需要钻井，矫直，并整理in sequence. Type 2 parts require only drilling and在序列。2型部分只需要钻finishing. The frequency of arrival and the time to be整理。到达频率和时间是routed to the drilling area are deterministic for both types路由到钻井区确定的两类of parts.零件。步骤1。确定问题。利用drills, straightener, and finishing operator needs to be演练，矫直机，完成运营需要assessed. In addition, the following modification to the评估。此外，以下修改的original system is of interest: the frequency of arrival of原系统的兴趣：到达的频率both parts is exponential with the same respective means这两部分是相同的手段指数as in the origi