Hedi Haddad

Using Cognitive Archetypes and Conceptual Graphs to Model Dynamic Phenomena in Spatial Environments

In this paper we propose a qualitative model to represent and reason about dynamic phenomena in a geographic space. Our model is based on linguistic cognitive archetypes, ontological definitions of geographic space and Conceptual Graphs (CGs). In a first part, we present the main concepts of the model and how we define them using CGs. In a second part, we present an overview of how this model is applied to the multiagent geosimulation domain in the context of the MAGS-COA project. Our model is original for two main reasons. First we use a linguistic approach to qualitatively model dynamic situations in a geographic environment. Second, we use CGs to represent the knowledge associated with such situations. Using CGs makes our model computationally feasible and useable to carry out spatial qualitative reasoning.

A framework to support qualitative reasoning about COAs in a dynamic spatial environment

We propose a framework to reason qualitatively about courses of action (COAs) which need to be executed in a realistic geographic space that may change. Particularly, the framework aims to support human mental ‘What-If’ analysis by simulating the execution of COAs in a virtual geographic environment, which can change during the simulation, and by allowing the user to explore various scenarios (different COAs) and to analyse their outcomes using causal reasoning techniques. In this article, we first present a framework which is based on a conceptual model of spatio-temporal situations, a multi-agent geosimulation platform and qualitative spatio-temporal causal reasoning techniques. Then, we illustrate the framework using a case study.

Using Multi-agent Geo-simulation Techniques for the Detection of Risky Areas for Trains

A transportation system is spatially and functionally distributed; its subsystems have a high degree of autonomy and are in constant interaction with each other and with the surrounding geographic environment. Modeling and simulating such systems in large-scale geographic spaces is a complex process. In this paper we address the domain of railway systems, and more particularly the problem of detecting risky areas along railroads. This requires that we consider a variety of static and dynamic variables, including train characteristics, hazardous events (e.g. rock-falls), and the properties of the large-scale geographic environment, as well as weather conditions. This simulation enables us to recommend speed limits in risky areas while taking into account all of the aforementioned factors. Since statistical and analytical models are not appropriate to represent such a complex process in which spatial constraints are of high importance, we adopted a multi-agent geo-simulation (MAGS) approach that facilitates the simulation of complex systems in large-scale geo-referenced environments. In this paper, we present Train-MAGS, an agent-based geo-simulation tool that simulates train behaviors in risky areas in large-scale virtual geographic environments. We also demonstrate how risky areas can be detected in real time using an agent-based approach. This work also illustrates how the application of artificial intelligence techniques, such as the MAGS approach, provides interesting perspectives of realistic and plausible simulations aimed at improving the functioning, the efficiency, and the safety of the transportation systems.

Using Causality Relationships for a Progressive Management of Hazardous Phenomena with Sensor Networks

Sensor networks prove extremely valuable in providing geo-information for any decision support system particularly those aiming to manage hazardous events. A thorough understanding and use of the semantics of this information allows for the identification and handling of impending hazardous events. An appropriate representation of the geo-information should boost this process. In this paper, we propose to encode causality relationships about natural phenomena and their effects in time and space with the concept of conceptual graphs. Using this encoding, we define the concepts of event and spatial propagation paths that enable the system to delimit the scope of sensed areas and use of sensing resources. These concepts also enable the system to set up priorities between the sensor network activities. These priorities are used to implement a progressive approach for the management of hazardous events.

A Knowledge-Based Multi-agent Geo-simulation Framework: Application to Intelligent Sensor Web Deployment

Sensor Web deployment is by nature a spatial problem since nodes are highly constrained by the geographic characteristics of the environment. Therefore, there is a need for an efficient modelling paradigm to address the issue of SW deployment taking into consideration the constraints of the geographic space. In this paper we propose a knowledge-based multi-agent geo-simulation framework to support the simulation of SW deployments in Informed Virtual Geographic Environments. This framework builds on our previous works on Informed Virtual Geographic Environments generation, on spatially reasoning agents and on qualitative reasoning about geo-simulation results. We illustrate the framework with a scenario of a sensor web deployment for weather monitoring purposes.

Sensor Network and GeoSimulation: Keystones for Spatial Decision Support Systems

Natural hazards and man-made disasters are victimizing large numbers of people and causing significant social and economical losses. By developing efficient Spatial Decision Support Systems (SDSS), managers will be efficiently assisted in identifying and managing impending hazards. This goal could not be reached without addressing significant challenges, including data collection, management, discovery, translation, integration, visualization, and communication. As an emergent technology, sensor networks have proven efficiency in providing geoinformation for any decision support system particularly those aiming to manage hazardous events. Thanks to their spatially distributed nature, these networks could be largely deployed to collect, analyze, and communicate valuable in-situ spatial information in a timely fashion. Since some decisions are expected to be taken onthe-fly, the right data must be collected by the right set of sensors at the right time. In addition to saving the limited resources of sensor networks, this will speed up the usability of data especially if this data is provided in the right format. In order to boost the decision support process, a thorough understanding and use of the semantics of available and collected heterogeneous data will obviously help to determine what data to use and how confident one can be in the results ultimately. An appropriate representation of the geoinformation should enhance this process. Data collected by sensors is often associated with spatial information, which makes it voluminous and difficult to assimilate by human being. In critical situations, the hazard manager has to work under pressure. Coping with such collected data is a demanding task and may increase the risk of human error. In this context, Geosimulation emerges as an efficient tool. Indeed, mapping the collected data into a simulated environment which takes into account the spatial dimension may dramatically help the hazard manager to easily visualize the correlation between data collected by sensors and the geospatial constraints. In this chapter, we first present fundamental concepts of SDSS and the most important challenges related to their development. Second, we outline the sensor network technology as an emergent tool for leveraging SDSS. Third, we present the Geosimulation approach as another keystone to enhance SDSS. In this part, we summarize the current opportunities, research challenges, and potential benefits of this technique in SDSS. Finally, for better efficiency, we propose an encoding that emphasizes the semantics of available data and

Multi-Agent Geosimulation in Support to Qualitative Spatio-Temporal Reasoning: COAs’ “What if” Analysis as an Example

Multi-Agent Geosimulation (MAGS) is a relatively novel approach to model-building and application in the geographic sciences and geocomputing (Torrens, 2008). It is mainly characterized by the use of Agent-Based Models – particularly Multi-Agent Systems (MAS) and Geographic Information Systems (GIS) in order to model, simulate and study complex phenomena taking place in geographical environments (Benenson and Torrens, 2004; Moulin et al., 2003). Recent research works in MAGS focused on two main trends. The first trend consists in improving different conceptual and computational aspects of MAGS models such as development methodologies (Ali, 2008), 2D and 3D virtual geographic environments models (Silva et al., 2008; Paris et al., 2009), agents perception and navigation models (Silva et al., 2008), generic MAGS platforms (Blecic et al., 2008) and models calibration and validation (Hagen-Zanker and Martens, 2008). The second trend consists in applying MAGS techniques to solve new problems such as parking policies evaluation (Benenson et al., 2007), prediction of house prices evolution (Bossomaier et al., 2007) and public health risk management (Bouden et al., 2008), to mention a few. Although these works allow modeling and simulating several geospatial phenomena, they do not guarantee that the simulation results will be well understood by a human user. In fact, results of geosimulations are usually presented using statistical, mathematical and / or graphical techniques (Ali et al., 2007). The complexity of the simulated phenomena and the huge volume of generated data make these techniques difficult to be interpreted by users. Indeed, human reasoning is mainly qualitative and not quantitative. Therefore, we believe in the importance of linking MAGS models with qualitative reasoning techniques, and we think that this link will allow the development of new systems which support qualitative reasoning in spatial contexts. While some recent works have been interested in this issue (Furtao and Vasconcelos, 2007), to our knowledge there is a lack of works that address its theoretical and computational aspects. Our contribution in this chapter aims at proposing an approach that uses MAGS techniques to support qualitative spatio-temporal reasoning. Particularly, we are interested in supporting a specific kind of qualitative reasoning called “What-if” reasoning and its particular application to the planning of courses of actions 11