شبیه سازی جریان کار برای حمایت از تصمیم گیری عملیاتی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|21832||2009||17 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Data & Knowledge Engineering, Volume 68, Issue 9, September 2009, Pages 834–850
Simulation is widely used as a tool for analyzing business processes but is mostly focused on examining abstract steady-state situations. Such analyses are helpful for the initial design of a business process but are less suitable for operational decision making and continuous improvement. Here we describe a simulation system for operational decision support in the context of workflow management. To do this we exploit not only the workflow’s design, but also use logged data describing the system’s observed historic behavior, and incorporate information extracted about the current state of the workflow. Making use of actual data capturing the current state and historic information allows our simulations to accurately predict potential near-future behaviors for different scenarios. The approach is supported by a practical toolset which combines and extends the workflow management system YAWL and the process mining framework ProM.
Business process simulation is a powerful tool for process analysis and improvement. One of the main challenges is to create simulation models that accurately reflect the real-world process of interest. Moreover, we do not want to use simulation just for answering strategic questions but also for tactical and even operational decision making. To achieve this, different sources of simulation-relevant information need to be leveraged. In this paper, we present a new way of creating a simulation model for a business process supported by a workflow management system, in which we integrate design, historic, and state information. Fig. 1 illustrates our approach. We consider the setting of a workflow system that supports some real-world process based on a workflow and organizational model. Note that the workflow and organizational models have been designed before enactment and are used for the configuration of the workflow system. During the enactment of the process, the performed activities are recorded in event logs. An event log records events related to the offering, start, and completion of work items, e.g., an event may be ‘Mary completes the approval activity for insurance claim XY160598 at 16.05 on Monday 21-1-2008’.The right-hand side of Fig. 1 is concerned with enactment using a workflow system while the left-hand side focuses on analysis using simulation. In order to link enactment and simulation we use three types of information readily available in workflow systems to create and initialize the simulation model. • Design information. The workflow system has been configured based on an explicit process model describing control and data flows. Moreover, the workflow system uses organizational data, e.g., information about users, roles, groups, etc. • Historic information. The workflow system records all events that take place in ‘event logs’ from which the complete history of the process can be reconstructed. By analyzing historic data, probability distributions for workflow events and their timing can be extracted. • State information. At any point in time, the workflow process is in a particular state. The current state of each process instance is known and can be used to initialize the simulation model. Note that this current state information includes the control-flow state (i.e., ‘tokens’ in the process model), case data, and resource data (e.g., resource availability). By merging the above information into a simulation model, it is possible to construct an accurate model based on observed behavior rather than a manually-constructed model which approximates the workflow’s anticipated behavior. Moreover, the state information supports a ‘fast-forward’ capability, in which simulation can be used to explore different scenarios with respect to their effect in the near-future. In this way, simulation can be used for operational decision making. Based on this approach, the system design in Fig. 1 allows different simulation experiments to be conducted. For the ‘as-is’ situation, the simulated and real-world processes should overlap as much as possible, i.e., the two process ‘clouds’ in Fig. 1 need to coincide. For the ‘to-be’ situation, the observed differences between the simulated and real-world processes can be explored and quantified. In our implementation we ensure that the simulation logs have the same format as the event logs recorded by the workflow system. In this way we can use the same tools to analyze both simulated and real-world processes. To do this, we need state-of-the art process mining techniques to analyze the simulation and event logs and to generate the simulation model. To demonstrate the applicability of our approach, we have implemented the system shown in Fig. 1 using ProM  and YAWL . YAWL is a workflow management system that, as reported in this paper, has been extended to provide high-quality design, historic, and state information. The process mining framework ProM has been extended to merge the three types of information into a single simulation model. Moreover, ProM is also used to analyze and compare the logs in various ways. In  three common pitfalls in current simulation approaches were presented. (1) modeling from scratch rather than using existing artifacts, which leads to mistakes and unnecessary work, (2) focus on design rather than operational decision making, which is helpful for the initial design of a business process but less suitable for operational decision making and continuous improvement, (3) insufficient modeling of resources, i.e., the behavior or resources is typically modeled in a rather naïve manner. This paper addresses the first two pitfalls. While addressing the third problem is a challenging research topic in itself , we concentrate here on the first two problems. That is, we integrate existing artifacts that can be extracted from a workflow system into a ready-to-use simulation model, and we incorporate the current state of the workflow system in our simulation model to enable short-term simulation. This paper extends our previous work , in that we go into more detail about the architecture of the realized system, describe the generated simulation models and how they can load a specified initial state more closely, and present a new XML file format for workflow states that enables other workflow systems to interface with our tools in a standardized way. The paper is organized as follows. Related work is reviewed in Section 2. Section 3 describes the approach proposed. Section 4 presents a running example, which is then used in Section 5 to explain the implementation realized using YAWL and ProM. Section 6 describes our approach to incorporate state information in more detail and presents the new XML file format for workflow states. Section 7 concludes the paper by discussing the three main innovations presented in this paper.
نتیجه گیری انگلیسی
7. Discussion In this paper we presented an innovative way to link workflow systems, simulation, and process mining. By combining these ingredients it becomes possible to analyze and improve business processes in a consistent way. The approach is feasible, as demonstrated by our implementation using YAWL and ProM. To conclude, we would like to discuss the three main challenges that have been addressed in this research. 7.1. Faithful simulation models Although the principle of simulation is easy to grasp, it takes time and expertise to build a good simulation model. In practice, simulation models are often flawed because of incorrect input data and a naı¨ve representation of reality. In most simulation models it is assumed that resources are completely dedicated to the simulated processes and are eager to start working on newly arriving cases. In reality this is not the case and as a result the simulation model fails to capture the behavior of resources accurately. Moreover, in manually-constructed models steps in the processes are often forgotten. Hence simulation models are usually too optimistic and describe a behavior quite different from reality. To compensate for this, artificial delays are added to the model to calibrate it and as a result its predictive value and trustworthiness are limited. In the context of workflow systems, this can be partly circumvented by using the workflow design (the process as it is enforced by the system) and historic data. The approach presented in this paper allows for a direct coupling of the real process and the simulation model. However, the generated CPN models in this paper can be improved by a better modeling of resource behavior. Furthermore, this resource behavior needs to be approximated in some way. Here, the mining of historic data can help to automatically choose suitable simulation parameters. As a consequence, more advanced process mining algorithms that extract characteristic properties of resources are needed to create truly faithful simulation models. 7.2. Short-term simulation Although most workflow management systems offer a simulation component, simulation is rarely used for operational decision making and process improvement. One of the reasons is the inability of traditional tools to capture the real process (see above). However, another, perhaps more important, reason is that existing simulation tools aim at strategic decision making. Existing simulation models start in an arbitrary initial state (without any cases in the pipeline) and then simulate the process for a long period to make statements about the steady-state behavior. However, this steady-state behavior does not exist (the environment of the process changes continuously) and is thus considered irrelevant by the manager. Moreover, the really interesting questions are related to the near-future. Therefore, the ‘fast-forward button’ provided by short-term simulation is a more useful option. Because of the use of the current state and historic data, the predictions are more reliable and valuable, i.e., of higher quality and easier to interpret and apply. The approach and toolset presented in this paper enable short-term simulation. A drawback is that in the current implementation three different systems are used. For example, the translation of insights from simulation via ProM and CPN Tools to concrete actions in the workflow system YAWL can be improved. Further research is needed to provide a seamless, but generic, integration. An interesting question regarding short-term simulation is how long this “short-term” can actually be. In general, the time horizon of interest depends on the questions that people have. However, assuming that a business process owner has a short-term simulation tool at hand, one also needs to consider the delay of decisions, or the delay of the realization of decisions, which has an impact on the estimated values in the predicted interval. 7.3. Viewing real and simulated processes in a unified manner Both simulation tools and management information systems (e.g., BI tools) present information about processes. It is remarkable that, although both are typically used to analyze the same process, the results are presented in completely different ways using completely different tools. This may be explained by the fact that for a simulated process different data is available than for the real-world process. However, the emergence of process mining techniques allows for a unification of both views. Process mining can be used to extract much more detailed and dynamic data from processes than traditional data warehousing and business intelligence tools. Moreover, it is easy to extend simulation tools with the ability to record event data similar to the real-life process. Hence, process mining can be used to view both simulated and real processes. As a result, it is easier to both compare and to interpret ‘what-if’ scenarios. Finally – while a detailed evaluation of the generated simulation models is beyond the scope of this paper – a unified view of real-life logs and simulation logs enables the validation of the simulation model by re-analyzing the simulation logs in a ‘second pass’ . This way, we can ensure that the ‘as-is’ situation is captured appropriately by the simulation model (by comparing process run times, availabilities, etc.) before starting to analyze ‘what-if’ scenarios. It can be anticipated that further topics become relevant for operational decision support systems in the future. In particular, for dynamic processes additional challenges might emerge. For example, by then additional runtime information (e.g. change logs, application context data) might exist, which, in turn, raises additional challenges with respect to the merging of different build- and runtime information in one simulation model.