A Proposed Model for Generic Agent-Based Simulation - College of ...

The 3rd National Graduate Conference (NatGrad2015), Universiti Tenaga Nasional, Putrajaya Campus, 8-9 April 2015.

A Proposed Model for Generic Agent-Based Simulation Loo Yim Ling College of Information Technology, Universiti Tenaga Nasional, Putrajaya Campus, 43000 Kajang, Selangor Darul Ehsan, Malaysia. Email: [email protected] Alicia Y. C. Tang College of Information Technology, Universiti Tenaga Nasional, Putrajaya Campus, 43000 Kajang, Selangor Darul Ehsan, Malaysia. Email: [email protected] Azhana Ahmad College of Graduate Studies, Universiti Tenaga Nasional, Putrajaya Campus, 43000 Kajang, Selangor Darul Ehsan, Malaysia. Email: [email protected] Abstract—Agent-based modeling and simulation had been a simulation approach implemented in application domains such as economy, infrastructure and social sciences since the past decade. The approach was implemented mostly in ad-hoc or domainspecific basis which caused a number of limitations. For instance, the nature of ad-hoc or domain-specific is meant to be used by a single research due to propriety issue. This caused the simulation model was not able to be reused in other research in similar application domain and validation of simulation results through replication could not be done. This paper proposed a model of generic agent-based simulation, which addresses to the limitations that caused handicaps for the past research works. Keywords—agent-based simulation; generic; agent-based simulation model

I.

INTRODUCTION

Simulation has been an emerging approach for understanding and predicting phenomenon of different domains, such as social science, business, education, economy and more [1]. Robustness and reliability of simulation approach in different areas had resulted in the emergence of different approaches for executing simulations. The approaches of simulation emerged from cellular automata to agent-based modeling, which had been recognized as a reliable modeling approach for simulation that imitates real life situations [2]. Agent-based modeling became well-known for the ability of the modeling approach to accommodate dynamics of a simulated application domain which is essential for real life simulation. Agents, which represents individuals that interact and learn, instead of objects, is the main role that enable agent-based modeling approach to respond to the dynamics. Boids simulation [2] is an example of the ability of an agent-based model, responding to dynamics of a simulated application domain.

Fig. 1. Boids Simulation [2]

The simulation illustrated in Fig. 1, shows the capability of interactive agents responding to dynamic environment by evolving into a certain behaviour with initial simple rules. Because of the robustness of agent-based modeling in simulation, rising number of research that adopt agent-based modeling for simulations of different domains were found such as business [4,6], transportation [5], environmental [9] and social sciences [7,10]. However, the modeling approach was much implemented in ad-hoc basis to individual domains without proper or no documentation at all. This caused continuous ad-hoc development of new agent-based simulation model, for reference of information for existing models were not released due to propriety issue and different models were designed to meet the specific objectives of the simulation. This caused large consumption of time and money, to develop from scratch for every new agent-based simulation research, but also draw limitation for validation of reliability of simulation results

This work is sponsored by Kementerian Pengajian Tinggi, Malaysia under the Fundamental Research Grant Scheme (FRGS).

ISBN 978-967-5770-63-0

327


and robustness of the simulation model [1,3]. Apart from causing large consumptions of resources, ad-hoc development of agent-based simulation model also caused inability of extensive research work on crucial phases such as; result validation and verification to ensure the reliability of results and robustness of simulation model [6,9] and extensibility of the model to accommodate simulations of similar application domain [4,5,6,9,10]. Further details of current research works that attempted to resolve these limitations of the existing agent-based simulation models will be discussed in Section II. The initiations of the current research works gave insights and inspirations towards proposal of a new model for generic agent-based simulation. Hence, the proposed model for generic agent-based simulation will be discussed in Section III of this paper. Section IV discusses the conclusions and further works to be implemented based on the proposed conceptual model. II.

EMERGENCE OF RECENT AGENT-BASED MODELING AND SIMULATION

Agent-based modeling approach for simulation systems had been emerging in attempt to have a better simulation model. Research works of agent-based simulation models in the past inherit different limitations, as mentioned in the earlier section of this paper. A few research works attempted to overcome the need of developing new simulation model in every new research that reside in similar domain. Instead of ad-hoc or individual domain agent-based simulation models, the research develop models which are generic enough to accommodate simulations of different specific domains within the main domain. For instance, the recent research by Zutshi et al. [4] developed a generic digital business simulation model. TransiTUM [5] expanded the idea of generic simulation model, where an unfixed model was constructed in order for further customization. FMS, a generic model for agent-based financial markets simulations [6] addressed that generic model by nature, is to be reused by others and thus has to be validated and verified. An agent-based generic model for human-like ambience [7] addressed to the area of replication of simulation result, a research phase that has virtually never been done but must be considered [1]. Schreinemachers and Berger [8] addressed that the main bridge to understanding and reuse of a developed simulation model lies within the documentation. However, to date, the use of a standardize documentation for documenting agent-based simulation model is still rare.

A list of the major limitations found in the previous agentbased simulation model and initiation of different researchers in attempt to resolve the limitations, is tabulated in Table I. TABLE I. PAST MODELS LIMITATIONS AND RECENT RESEARCH EMERGENCE Limitations of past research Ad-hoc or domainspecific model Inextensible customizable

or

Lack of result validation and verification Inflexibility of replication and reuse

Attempts of resolving the limitations Zutshi et al. [4], Bidermann et al. [5], Bagneris [6], Hennicker et al. [9], Luo et al. [10] Biedermann [5], Zutshi et al. [4], Bagneris [6], Hennicker et al. [9], Luo et al. [10] Bagneris [6], Hennicker et al. [9]

Schreinemachers and Berger [8], Hennicker et al. [9], Bosse et al. [7]

Due to the ad-hoc or domain specific development of simulation models in the past, much resources such as time and money had been wasted to develop fresh agent-based simulation models and systems from scratch. The research done by Zutshi et al. [4] addressed the issue by development of a generic digital business model that could accommodate different digital business models instead of one. This would enable specific business models not to repetitively develop adhoc simulation models, but reuse the existing generic model which would save time and cost. The model was validated to be generic enough to support different digital business models through case study simulation of both Custo Justo and Facebook business models. The practice of the past research did not end in causing large consumption of resources alone, but the nature of having propriety issue for the past researches caused the simulation models developed to be inextensible and not customizable. The lack of release of documentations for the simulation models caused other researchers to have no way of extending or customizing the existing model to meet the objective of their individual research, even though they have found similar models to be applied. This is another issue that will be discussed further in the final part of this section. Thus, in the research done by Biedermann et al. [5] the issue of extensibility and further customization was addressed for multiscale pedestrian models was addressed. This research enabled extension or customization to be able to be applied on the model developed, in order for the model to not only simulate for one function of multiscale coupling of pedestrian simulation, but a few multiscale couplings. This attempt had achieved resolution in another area of initiation of a generic model, which is, instead of addressing limited simulation domains, the

ISBN 978-967-5770-63-0

328


model addressed limited simulation functions within the domain. The same goes to the research of generic model for modeling human-like behaviours in crowd simulation by Luo et al. [10] where the model was aimed to be further customized and used by other researchers, following the dynamic change of human behaviours. The research works enabled new research insights to reuse existing research model with some tweaking of configurations to meet the specific requirements of the domain specific simulation, instead of having to develop the simulation model from scratch. As the model was made available to be altered by own author or other researchers that are interested to extend or customize the model to meet their specific research needs, it is important that the model is validated and verified. Research works done by Bagneris [6] as well as Hennicker et al. [9] addressed to the issue of having the generic simulation model validated and verified before opening the access of the model to others for further enhancements and customizations. However, there is still lacking of standardized process and methodology of validation and verification [1,3], where different research works used different method of validation, causing the validation and verification process to be questionable. For instance, all the research works mentioned in Table 1, practiced total different validation and verification methodologies. Therefore, the attempt to resolve this limitation is still in enhancement and need further improvements. Whenever validation and verification of a system is mentioned, replication of results and reuse issues are recommended [1]. Replication of results enable testing of reliability of simulated result and the robustness of the simulation model, through the generation of results using different data, yet same environment. For instance, a research done by Bosse et al. [7] addressed to the limitation through opening the access to the simulation model to public and whoever that used the model can feedback whether the model does give reliable simulation results. Unfortunately, replication and reuse is still not a widely practiced stage of research process, as ad-hoc and domain-specific research methodology is still being practiced and generic models researches have yet to embrace proper documentation for others to study and understand the existing model for replication. For instance, the standardized documentation using Overview, Design Concepts and Details namely, ODD Protocol [11] was not used in either of the research works mentioned in Table I. This imply that different ways of definitions and explanations for the simulation model are done, since the standardized documentation was not used. A research done by Schreinemachers and Berger [8] was criticized not to have a

standardized documentation although the simulation model was clearly defined and documented according to the authors’ belief. The authors then took the critic and initiated drafting out the simulation model using ODD Protocol, a standard documentation for agent-based simulation models as addressed by the critic as well. Within the documentation, Schreinemachers and Berger [8] addressed that the standardized documentation namely, ODD Protocol is essential, if the simulation model is to be made understood by others and eventually be replicated and reused. However, this is again still in emergence as the field is still new and rare in application by other research works. Nonetheless, the emergence of agent-based modeling and simulation technology had inspired and motivated to the idea of having a model for generic agent-based simulation that address to the issues mentioned above. The section below discuss both the insights and the motivation in detail. III.

A PROPOSED MODEL FOR GENERIC AGENTBASED SIMULATION

Agent-based modeling had been an emerging approach in research for being implemented for simulation systems. The emergence found from the researches done, had given great insights and motivation for the proposal of a simulation model that address to the limitations that had been tried to be resolved in vain. A Generic Model for Agent Based Simulation Component 1 : Repository Generate knowledge base

Component 2 : Simulation Generate algorithm Execute simulation

Component 3 :Validation Validate results

Component 4 : Replication Replicate results

Fig 2. Conceptual model of generic agent-based simulation

Fig. 2, illustrates the conceptual architecture of the model designed for generic agent-based simulation. Within the

ISBN 978-967-5770-63-0

329


conceptual model, there are four main components, namely repository, simulation, validation and replication. In repository component, data are collected and stored into a knowledge base, acting as the brain of the simulation as suggested in the research done by Bagneris [6]. Conceptually, the data collected shall consist of information from at least three different domains to ensure the algorithm generated in the next component will be generic enough to be implemented to at least three different domains. This imply that the knowledge base determine the algorithm that will be generated that governs the simulation in the next process. Previous works on generic agent-based simulation models focused on simulations of specific environments that belong to the same domain e.g. Zutshi et. al [4] for simulations of Custo Justo and Facebook environments but within the same domain of business models. However the conceptual model of this generic agent-based simulation aims for cross domain genericity. Thus, data collection and extraction will consist of three different domains e.g. transportation, business and agriculture. Common parameters or characteristics of the three different domains will have to be extracted from the collected data which resides in the knowledge base. These common characteristics will shape the generation of generic algorithm, which will govern the simulation. In this conceptual simulation model, the more common characteristics captured, the more the genericity of the initial generic algorithm generated. Second component which is namely simulation, consists of two main entities which are, generic algorithm and simulation engine. The generic algorithm generated from the knowledge base is aimed to be able to govern the simulation to simulate phenomenon of the three different domains. The generic algorithm will be the neuron system which configures and determines the parameters to be used in a particular requested simulation. Simulation is executed through implementing the generic algorithm to generate requested simulation results. After simulation results are acquired, the results need to be validated through validation component. Within validation component, simulation results are aimed to be validated of its reliability through comparing the simulation results with collected data in knowledge base just as suggested by the research done by Zutshi et al. [4]. If the results validated to be ambiguous, further customization need to be applied on the generated generic algorithm until reliable simulation results are acquired. However, if the results were validated to be reliable, the simulation model is proven of its generic nature for the three domains simulated. Then the next process of replication and reuse will be followed.

Replication component is the stage of research where Bosse et al. [7] firmly addressed. The aim of replication component in this model is to make other researchers understand the model and use it to generate simulation results for their own purpose with their own data. The domain of the other researchers that use the simulation model need to be similar to the three domain choices and the results generated is expected to be nonambiguous. This will further ensure the reliability of the simulation results and robustness of the simulation model to be implemented across the domains, with not just the existing knowledge base. Before the other researchers able to use the simulation model, they need to first understand the model and simulation system. Thus, the documentation of the simulation model is crucial and the standardized ODD Protocol is aimed to be used for this research as suggested in the research done by Schreinemachers and Berger [8]. This will not only enable unambiguous definition of the simulation model but also standardize the definitions and terminologies of the simulation model which is not practiced in ad-hoc or domain specific simulation model research [3]. With the conceptual architecture developed, the model for generic agent-based simulation is expected to be able to accommodate simulation of at least three different domains. It is an attempt to resolve the issues which the emerging agentbased modeling and simulation technology is trying to resolve as well. IV.

CONCLUSION AND FURTHER WORKS

This paper presents the limitations found in the current agentbased modeling and simulation research works. The ad-hoc and domain-specific, inextensible, lack of result validation and verification as well as inflexibility of replication basis of simulation model development, have put quite a handicap for extensive research works to be done to the current research works. As a result, new ad-hoc basis of research works are found more than extensive development of existing research work. Thus, a model for generic agent-based simulation is proposed in this paper which consists of four different components. Each components are geared to address to the limitations and aimed to resolve the issues. The conceptual simulation model developed in this research as a result of studies across different agent-based modeling and simulation research is still in the process of refinement. The components in the proposed model of generic agent-based simulation have to be further customized of the methodologies

ISBN 978-967-5770-63-0

330


and parameters to be used and included. Thus, the research will continue on with refining each of the components in detail. REFERENCES [1]

Axelrod, R. (1997). Advancing the art of simulation in the social sciences. In Simulating social phenomena (pp. 21-40). Springer Berlin Heidelberg. [2] Macal, C. M., & North, M. J. (2009, December). Agent-based modeling and simulation. In Winter Simulation Conference (pp. 86-98). Winter Simulation Conference. [3] Heath, B., Hill, R., & Ciarallo, F. (2009). A survey of agent-based modeling practices (January 1998 to July 2008). Journal of Artificial Societies and Social Simulation, 12(4), 9. [4] Zutshi, A., Grilo, A., & Jardim-Gonçalves, R. (2014). A Dynamic Agent-Based Modeling Framework for Digital Business Models: Applications to Facebook and a Popular Portuguese Online Classifieds Website. In Digital Enterprise Design & Management (pp. 105-117). Springer International Publishing. [5] Biedermann, D. H., Kielar, P. M., Handel, O., & Borrmann, A. (2014). Towards TransiTUM: A generic framework for multiscale coupling of pedestrian simulation models based on transition zones. Transportation Research Procedia, 2, 495-500. [6] Bagneris, J. C. (2012). FMS, a Generic Framework for Agent-Based Financial Markets Simulations. Available at SSRN 2149543. [7] Bosse, T., Hoogendoorn, M., Klein, M. C., & Treur, J. (2008). An agentbased generic model for human-like ambience. In Constructing Ambient Intelligence (pp. 93-103). Springer Berlin Heidelberg. [8] Schreinemachers, P., & Berger, T. (2011). An agent-based simulation model of human–environment interactions in agricultural systems. Environmental Modelling & Software, 26(7), 845-859. [9] Hennicker, R., Bauer, S., Janisch, S., & Ludwig, M. (2010, July). A generic framework for multi-disciplinary environmental modelling. In Fifth Conference of the International Environmental Modelling and Software Society, Ottawa, Canada (pp. 980-994). [10] Luo, L., Zhou, S., Cai, W., Low, M. Y. H., & Lees, M. (2009, September). Toward a generic framework for modeling human behaviors in crowd simulation. In Proceedings of the 2009 IEEE/WIC/ACM International Joint Conference on Web Intelligence and Intelligent Agent TechnologyVolume 02 (pp. 275-278). IEEE Computer Society. [11] Grimm, V., Berger, U., DeAngelis, D. L., Polhill, J. G., Giske, J., & Railsback, S. F. (2010). The ODD protocol: a review and first update. Ecological modelling, 221(23), 2760-2768.

ISBN 978-967-5770-63-0

331