روش مدل سازی آبشار بحران بر اساس ادغام شبکه بیزی
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|29304||2014||12 صفحه PDF||سفارش دهید||9102 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : International Journal of Approximate Reasoning, Volume 62, June 2014, Pages 94–105
This paper presents a Bayesian Network (BN)-based modeling method for cascading crisis events. Crisis events have occurred more frequently in recent years, such as typhoons, rainstorms, and floods, posing a great threat to humans. Addressing these crises requires a more effective method for crisis early-warning and disaster mitigation in crisis management. However, few modeling methods can combine the crisis chain reaction (macro-view) and the elements within the crisis event (micro-view) in a cascading crisis events. Existing classical methods fail to consider the causal relations linking the micro to macro level in crisis events, which affects the forecasting accuracy and effectiveness. Based on systems theory, this paper first abstracts the crisis event as a three-layer structure model consisting of input elements, state elements and output elements from a micro-view. Next, a cascading crisis events Bayesian Network (CCEBN) model is developed by merging the single crisis events Bayesian Networks (SCEBNs). This method efficiently combines the crisis event's micro-view and the macro-view. The proposed BN-based model makes it possible to forecast and analyze the chain reaction path and the potential losses due to a crisis event. Finally, sample application is provided to illustrate the utility of the model. The experimental results indicate that the method can effectively improve the forecasting accuracy.
A well-performed emergency decision support system (EDSS) plays a crucial role in emergency-warning and disaster mitigation of crisis management. To save lives and prevent additional property damage, several models for EDSS have been developed in the field of crisis management. However, these models did not comprehensively utilize the information of the whole crisis environment, which was restricted to be specific to its own context. Most of the former studies focused only on a single crisis event (such as the artificial neural network-based EDSS , nuclear emergencies , hurricanes in Florida , floods in Italy , among others). These methods overlook the interactions among different crisis events, which results in a low forecasting accuracy. When a crisis event occurs, it usually leads to secondary events or derived events. Specifically, there are causal relations among the different crisis events. The occurrence and development of one event always has impact on the other events. This phenomenon can be called chain reaction of crisis events. These corresponding associated crisis events form a cascading crisis event. It is helpful to improve the effectiveness of crisis management by understanding a crisis event and identifying its cascading event comprehensively as well as accounting for the chain reaction information. Several methods include this principle, but most of these studies take the crisis event as a node with a macro-view, which cannot reveal the evolution mechanism of the cascading crisis events completely. For example, some studies focused on the qualitative analysis of a domain-specific cascading crisis event, such as a geological-hazard chain  and , sediment-hazard chain , disaster mitigation framework , and typhoon-hazard chain . These studies especially focused on researching the interior mechanisms and manifestations of domain-specific cascading crisis events qualitatively but did not provide any modeling methods for the cascading crisis events. Other researchers focused on modeling the chain effects of the cascading events. J. T. Rodriguez et al.  proposed an alternative data-base approach to assessing the potential damage that can arise from various combinations of phenomena and locations. However, this method will result in too many rules to model the complexity and uncertainty of the problems. C. Fang et al.  proposed a simulation-based risk network model for decision support in project risk management. This method accounted for the phenomena of chain reactions and loops, but it neglected the detailed connections of information among the internal components of a cascading crisis event. J.W. Wang et al.  studied the network model of the chain reaction based on complex network theory. They analyzed the topological features of the network from only a macroscopic perspective. However, all these methods could not consider both the crisis chain reaction (macro-view) and the elements within the crisis event (micro-view) in a cascading crisis event. Therefore, to make effective use of the chain reaction information and to reveal the evolutionary mechanism of the crisis event, it is important to model the cascading events by combining a single crisis event from a micro-view with the associated crisis events from a macro-view. This paper proposes a novel Bayesian Network (BN)-based modeling method that can combine the information that comes from the internal crisis events and the chain reaction process. In contrast to the traditional macro analysis methods, this paper focuses on the evolutionary mechanisms of crisis events and combines the microscopic and macroscopic ways of analyzing the chain reaction path and predicting the potential losses of the crisis events. An example of a typhoon–rainstorm–flood disaster chain is taken to demonstrate the validity of the proposed method. The remainder of this paper is structured as follows. Section 2 discusses why BN is used as the modeling tool. In Section 3, a single crisis event Bayesian Network model (SCEBN) is constructed. Section 4 proposes the cascading crisis events Bayesian Network (CCEBN). And an implementation of the CCEBN-based EDSS is presented in Section 5. Section 6 presents an experimental example to evaluate the model. Finally, in Section 7, we conclude with a summary and a statement concerning possible future research.
نتیجه گیری انگلیسی
This paper has presented a BN-based cascading crisis events modeling method, which is used to forecast and analyze the chain cascading crisis events. And an emergency decision support system (EDSS) based on the cascading crisis events Bayesian Network (CCEBN) model has been developed to support real-time decisions. The proposed model addresses the limitations of current methods, which cannot combine the elements within the crisis event (micro-view) with the crisis chain reaction (macro-view) in a cascading crisis event. Compared with the single crisis management, this proposed method can offer a more comprehensive emergency pre-warning. It is helpful to improve the ability of crisis management, such as disaster prevention and mitigation. The model is suitable for the crisis events which can be abstracted as a system that includes input elements, state elements and output elements. Furthermore, the data set of the crisis events should be acquired and discretized because the BN's ability to address continuous data is limited. From the micro-view, this paper abstracts the crisis event as a generic three-layer structure model that covers input elements, state elements and output elements that are based on systems theory. Additionally, based on the generic three-layer model, the paper proposed a modeling method for the single crisis event Bayesian network; it is a fundamental model for a combination of single crisis event Bayesian networks in a cascading crisis event. This paper simplifies the interrelationships among the crisis events through judging whether there is an intersection between the output elements and input elements from two different crisis events. Additionally, a modeling method of CCEBN is proposed through merging the crisis event Bayesian network. It could comprehensively consider the influence from both the elements of the crisis event itself and the associated crisis events in a cascading crisis scenario. Thus, compare with the conventional methods that take the whole event as a research object with a macro-view, this study is also focused on the internal reaction mechanism among crisis events in a cascading crisis event. Thus, the model can not only utilize the chain reaction information among the crisis events but also take advantage of the powerful reasoning capacity of the Bayesian network. Once we have collected sufficient historical data, the crisis management can be achieved by forecasting the evolutionary path and overall loss of the crisis events based on the current data. The selected experimental case study analyzes typical cascading crisis events that are composed of a typhoon event and the secondary flood event and rainstorm event that occurred in Taiwan during 2000–2007. Additionally, the experimental results show that the cascading crisis event Bayesian Network proposed in this paper could improve the forecast accuracy compared with the single crisis event Bayesian Network because the model comprehensively considers the influence from both the crisis event itself and the related crisis events in the cascading crisis scenario. There are some limitations and potential extensions of the proposed method. BN demands oriented links, is inherently acyclic , and hence does not easily model the loop phenomenon. To overcome this limitation, temporal or spatial dynamics can be modeled in BNs using a separate network for each time slice . And there is a chance that the error propagation of uncertainty may emerge in the BN which would limit the forecasting accuracy. To avoid this problem, the key is to improve the data quality. In addition, the problem also can be solved by model improvement or algorithm innovation. Such as Y. Zhu et al.  present an error back propagation algorithm to revise the error parameters during the propagation of uncertainty, which can be used to solve the limitation. Furthermore, the forecasting model is based on historical data. Thus, the data collection would be difficult to use for the model. However, with the development of the Internet and with increasing attention being paid to data, the problem of data collection becomes less of an obstacle. The cascading crisis events system is a complex and uncertain system. Thus, to achieve faster and more accurate crisis management, a comprehensive analysis also must consider the system dynamics, for which we plan to attempt to apply the dynamic Bayesian Network (DBN) to cascading crisis events modeling in the future.