融合企业建模与软件设计：第七届国际研讨会成果

需积分: 9 66 浏览量更新于2024-07-19 收藏 14.18MB PDF 举报

"《业务建模与软件设计：第七届国际研讨会》于2018年发表在Springer出版物中，该论文探讨了企业模型和软件设计在信息技术日益重要背景下的融合。四十年前的企业环境与现在有着显著差异，当时没有计算机支持的业务流程自动化存在诸多限制，信息交换主要依赖于非全球性的通信手段，多媒体技术也不发达。然而，随着互联网、Web服务和云计算基础设施的发展，企业逐渐对外部化业务流程和开发适应性强的信息系统变得更加依赖。论文强调了企业工程（EE）和软件工程（SE）作为独立学科的发展，但在当前，将企业建模与软件规格化结合起来至关重要。通过建立一个共同的企业-软件概念基础，可以创建抽象模型来捕获实体、过程和规则，无论是在软件开发还是企业环境中。这种统一的框架有助于实现企业-软件的一致性和可追溯性。然而，实现这一目标面临多方面的挑战。首先，从系统的角度来看，需要同时处理结构、动态、数据并保持整体一致性。其次，当考虑企业信息系统时，必须兼顾企业工程和软件工程的观点，确保软件与其业务环境的一致性。此外，企业/软件系统建模可能采用黑盒（功能）或白盒（操作）视角，但两者之间的模型必须相互一致。对于上下文感知的信息系统，需权衡内部流程优化和用户体验提升之间的平衡。最后，在软件规格化过程中，可能需要融入如隐私、透明度等公共价值观，这些通常是非功能性需求，需要转化为实际的系统功能。在第七届BMSD——业务建模与软件设计国际研讨会的讨论中，专家们着重讨论如何整合社会理论（企业建模的主要来源）和计算范式（软件规格化的根源），以实现有效的结合。论文参考了LNBI普雷弗（LNBIP 275），强调了促进企业工程师和软件工程师之间合作的重要性，以进一步联结业务建模和软件设计，这对于BMSD社区的发展具有关键意义。该研讨会于2017年7月3日至5日在西班牙巴塞罗那举行，收录了经过修订的精选论文，反映了当时对跨领域协作与研究的前沿思考。"

A Visionary Way to Novel Process Optimizations 3

concepts, the Sect. 3 derives objectives for a NPM, NPS and NPO. The Sect. 4

provides corresponding designs, followed by their demonstration and evaluation.

A Sect. 7 concludes the paper.

2 Underlying Concepts

Starting with the selection of a modeling approach and the question, how pro-

cesses can be simulated and optimized in the Subsect. 2.1, the Subsect. 2.2 refers

to underlying knowledge generation concepts. A further section introduces ANN.

2.1 Process Domain

Following the fundamental procedure model for simulation studies of Gronau [9],

a model creation is realized after the modeling purpose has been deﬁned, ana-

lyzed and corresponding data has been collected. Hence, the following starts with

modeling issues. Afterwards, as the model is valid, simulation studies are realized

and simulation results collected, analyzed and interpreted. As changes or opti-

mizations are required, adjustments are deﬁned and simulations tested until a

suﬃcient solution has been identiﬁed. Best solution options will be implemented,

of course. The following characterizes required ﬁelds of the process domain fol-

lowing this procedural logic.

Process Modeling. Starting from a basic deﬁnition of models, which refer to

simpler mappings of relevant attributes of the real world with the intention to

reduce the complexity of the real world with respect to modeling objectives,

process models can be understood as a homomorphous, time-based mapping

of a real-world system focusing a sequence-based, plausible visualization [9].

According to Krallmann et al. [21], a system to be modeled consists of an amount

of system elements, that are connected with an amount of system relations.As

it is limited by a system border,thesystem environment and the system are

connected with an interface to exchange system input and system output.

For the modeling of process systems, several process modeling languages

can be used. Considering organizational, behavior-oriented, informational and

knowledge-oriented perspectives, Sultanow et al. identify the Knowledge Model-

ing Description Language (KMDL) to be superior in comparison to twelve com-

mon modeling approaches [37]. Because of the analogy with a human brain as

knowledge processing unit, especially a knowledge process modeling is focused.

Here, Remus gives an overview of existing modeling methods and a comparison

of their ability to represent knowledge [28, p. 216f]. ARIS, EULE2, FORWISS,

INCOME, PROMOTE and WORKWARE are only some representatives. Again,

the KMDL can be identiﬁed to be superior because of its ability to overcome

lacks in visualizations and analyses through the combination of several views

such as the process view, activity view and communication view [14]. This lan-

guage has been developed over more than ten years. Having collected experiences

in numerous projects of various application areas such as software engineering,

4 M. Grum and N. Gronau

product development, quality assurance, investment and sales, the evolution of

the KMDL can be found in [12]. The current version refers to KMDL version

2.2, but the development of version 3.0 is in progress [10]. In addition to the

modeling language, the KMDL reaches a fully developed research method [11].

With its strengths in visualization and the focus of knowledge generation,

the KMDL seems attractive for a transfer to the neuronal level. To the best of

our knowledge, such a transfer has not been realized yet in any other process

modeling language. With its intention to focus on the generation of knowledge

following Nonaka and Takeuchi and the intention to transfer the learning poten-

tial of ANN, the KMDL enables the modeling of tacit knowledge bases and single

or numerous knowledge transfers beside common processing issues. Hence, the

KMDL is selected as modeling language for the demonstration in Sect. 5.The

current paper builds on the wide spread KMDL version 2.2 [14].

Process Simulation. Once, a valid process model has been created, a dynamic

sequence and variations of this process can be simulated. Aiming to gain insights

within a closed simulation system, the intention is to transfer insights to reality.

For this, the following pre-conditions have to be fulﬁlled: process models have to

provide completeness. This includes the registration of input data such as time,

costs, participants, etc. Further, process models have to provide interpretability

of decisions. Here, values of variables, state change conditions and transfer prob-

abilities are included. Further, meta information have to be considered, as for

example the number of process realizations within a simulation. Beneath further

objectives, the following simulation scenarios can be evaluated quickly and at

low costs: current sequences of operations, plans and process alternatives. Those

evaluations can be realized before expensive adjustments within current process

models (so called as-is models) are implemented in the real world [9].

Process Optimization. As processes are adjusted with the intention to opti-

mize them in regard to a certain objective, one speaks from process optimization.

All activities and decisions that lead to a desired optimization of business pro-

cesses, are designated as business process optimization [9]. The success of an

optimization is measured by key performance indicators, such as production

time, failure and success rates, produced components, etc. There can be found

two basic approaches for business process optimizations that are reﬂected in

various methods and variations:

(1) An approach called Continuous Improvement Process and

(2) an approach called Business Process Reengineering.

(1) Originally inspired by a Japanese living and working philosophy called

Kaizen, the management concept realizing a never ending improvement of pro-

cesses and products in small steps is referred to as Continuous Improvement

Process, (CIP). Following Imai, key principles are a feedback mentality and pro-

cesses are reﬂected continuously. The everlasting search for eﬃciency demands

A Visionary Way to Novel Process Optimizations 5

for the identiﬁcation and improvement of suboptimal processes, such that waste

is reduced and eliminated. Further, the emphasis lies on continual steps rather

than giant changes, which is connected to the key principle of evolution [18]. Cor-

responding management concepts follow cyclic procedures and can be found in

numerous variations: Shewhart Cycle [36], Deming Wheel [5], a second Shewhart

Cycle [6], PDSA Cycle [7], PDCA Cycle [23] and a second PDCA Cycle [19].

In general, those concepts contain planning activities (plan). Afterwards, the

process is implemented and carried out (do). A feedback is collected and com-

pared to a planned output (check ). Then changes are implemented constantly or

revised, before this cyclic procedure is started again. The feedback collection can

gather process data either coming from a process realization of the real world,

or from a simulated processes. Since improvement ideas are generated during

the process realization and a focus lays on single processes, improvements are

carried out bottom-up.

(2) The concept of Business Process Reengineering, (BPR), refers to the funda-

mental over-thinking of as-is processes [15]. This is mostly connected with far

reaching changes up to a completely new design of processes and the organiza-

tion itself. Process improvements are designed as if the organization was built

anew and current knowledge and state-of-the-art techniques are considered addi-

tionally. Here, improvements are carried out top-down and optimization results

provide the following characteristics: Decisions are decentralized, process step

sequences are reorganized and diﬀerent process variations can be considered.

Further, the localization of working content is organized meaningfully, the need

for control and required coordination eﬀorts are reduced and centralized contact

points (e.g. for customer requests) are established.

In conclusion, within the ﬁeld of process modeling, the modeling language

KMDL is suitable for the use as NPM. Since the KMDL does not provide simula-

tion structures, yet, ANN simulation capabilities can be used for an enhancement

of the KMDL and enable NPS. Further, learning capabilities of ANN shall be

used for the creation as NPM purposes and optimization as NPO purposes. Here,

a CIP is attractive for ﬁrst steps and a BPR can realize further potentials.

2.2 Knowledge Representation

Nonaka and Takeuchi distinguish between explicit knowledge and tacit knowl-

edge [24]. While the ﬁrst can be verbalized and externalized easily, the sec-

ond is hard to detect. The following four knowledge conversion types can be

distinguished:

–Aninternalization is the process of integration of explicit knowledge in tacit

knowledge. Here, experiences and aptitudes are integrated in existing mental

models.

–Asocialization is the process of experience exchange. Here, new tacit knowl-

edge such as common mental models or technical ability are created.

–Anexternalization is the process of articulation of tacit knowledge in explicit

concepts. Here, metaphors, analogies or models can serve to verbalize tacit

knowledge.

6 M. Grum and N. Gronau

–Acombination is the process of connection of available explicit knowledge,

such that a new explicit knowledge is created. Here, a reorganization, recon-

ﬁguration or restructuring can result in new explicit knowledge.

With the intention to focus on the potentials of human brains and its gener-

ation of knowledge, the knowledge generation concepts of Nonaka and Takeuchi

seem attractive for modeling on a neuronal level. Since the KMDL is the only

modeling language, which builds on this concept, the KMDL is selected for

demonstration purposes.

2.3 Artiﬁcial Neuronal Networks

Originally, neural networks were designed as mathematical models to copy the

functionality of biological brains. First researches were done by Rosenblatt [32],

Rumelhart et al. [33] and McCulloch and Pitts [22]. As the brain connects sev-

eral nerve cells, so called neurons, by synapses, those mathematical networks

are composed of several nodes, which are related by weighted connections. As

the real brain sends electrical activity typically as a series of sharp spikes, the

mathematical activation of a node represents the average ﬁring rate of these

spikes.

As human brains show very complex structures and are confronted with

diﬀerent types of learning tasks (unsupervised, supervised and reinforcement

learning), various kinds of networking structures have established, which all have

advantages for a certain learning task. There are for example Perceptrons [31],

Hopﬁeld Nets [17], Multilayer Perceptrons [1,33,38], Radial Basis Function Net-

works [2] and Kohonen maps [20]. Networks containing cyclic connections are

called feedbackward or recurrent networks.

The following focuses on Multilayer Perceptrons and recurrent networks

being confronted with supervised learning tasks. Here, input and output val-

ues are given and learning can be carried out in minimizing a diﬀerentiable error

function by adjusting the ANN’s weighted connections. For this, numerous gra-

dient descent methods can be used, such as backpropagation [1,27], PROP [29],

quickprop [8], conjugate gradients [16,35], L-BFGS [3], RTRL [30]andBPTT

[39]. As the weight adjustment can be interpreted as a small step in a direction

of optimization, the ﬁx step size can be varied to reduce great errors quickly. The

learning rate decay can be used to reduce small errors eﬃciently and a momen-

tum can be introduced to avoid local optima. In this stepwise optimization,

analogies to continuous process optimizations can be found (see Sect. 4.3).

Since neuronal networks model human brains and the knowledge of a certain

learning task, the following refers to neuronal networks as neuronal knowledge

models. Those represent a current state of knowledge, the capability to generate

new knowledge through their activation and interaction and the possibility to

transfer knowledge or further process relevant objects within process simulations.

A Visionary Way to Novel Process Optimizations 7

3 Objectives of an ANN Process Domain

As one assumes to have a given process model and one aims to consider a neu-

ronal network as a process participant’s knowledge model within the simulation

of that process model, the following objectives have to be considered coming

from a modeling side:

1. Neuronal knowledge models have to be integrated within existing process

models.

2. The same neuronal knowledge models have to be able to be integrated several

times within a process model.

3. Neuronal knowledge models have to be integrated within process simulations.

4. Modeled environmental factors (material such as non-material objects) have

to be integrated with considered knowledge models.

5. Outcomes (materialized such as non-materialized) of considered knowledge

models have to be considered within the process model.

Further, objectives have to be considered coming from a neuronal techniques

side:

6. Neuronal tasks have to be considered while neurons follow biological models.

This includes both the neuron’s everyday business and learning processes.

7. Parallel neuronal task realizations have to be considered within neuronal

networks.

8. Time-dependent neuronal behaviors have to be considered within neuronal

networks.

9. Sequential neuronal task realization have to be considered within neuronal

networks.

10. Diﬀerent levels of neuronal task abstractions have to be considered in the

neuronal process modeling and simulation.

11. Sensory information and knowledge ﬂows have to be considered within the

modeled neuronal network.

12. Actuator information and knowledge have to be considered as outcomes of

neuronal networks.

Each identiﬁed objective of those domains is relevant for the transfer of a process

modeling language and serves as input for the following sections.

4 Design of an ANN Process Domain

The visionary way to a novel process optimization is drawn with help of the

following subsections. First, a design for a neuronal process modeling is given.

Then, a neuronal process simulation design follows. Finally, the neuronal process

optimization is designed. All designs refer to the neuronal process circles of Fig. 2.

Explanations can be found in corresponding subsections.

剩余214页未读，继续阅读

WindStand

粉丝: 35
资源: 367

融合企业建模与软件设计：第七届国际研讨会成果

ERP信息化专业资料:SAP专业学习资料BW330_-_2005-Q1_-_Ltr_-_Business_Information_Warehouse_-_Modeling.pdf

Method_of_Modeling_Differential_Vias-mod-Iss2.2-Apr2-12.pdf

新手也能搭建的chargtp 4.0 的中文教程

Sybase.Modeling.NMap.Runtime

matlab中close1of

relation.relation("WITHIN")

AttributeError: 'Tensor' object has no attribute 'encode'

heavy transport

ModuleNotFoundError: No module named 'transformers.modeling_bert'

amos调节效应分析

最新资源