Estrategias de monitorización de incendios forestales mediante redes inalámbricas de sensores autónomas

Resumo

PhD Summary This chapter presents the research topic selected for this thesis, and its motivation. It justifies the choice, presents the objectives of this research, and introduces the structure of the thesis. Subsequently, the contributions that have been made during the development of this research are enumerated, along with the main conclusions that can be drawn from this work. Finally, the various lines of work that can be explored in the future will be described. 1.1 Motivation Worldwide, wildfires burn significant amounts of forest. Due to climate change and the evolution of land use (increasingly overexploited), forest fires are becoming more frequent and more severe [175] [107]. To cite an example, only in southern Europe, about half a million hectares are consumed each year [106]. However, the impact of forest fires is not limited solely to the amount of vegetation consumed. Forest fires also have significant environmental consequences (e.g. destruction of wildlife habitats), economic (e.g. modifying the assessed value of the property) and social (e.g. loss of lives, or damaging the development of tourism activities). 16 PhD Summary The amount of lives taken by forest fires deserves a special mention. Both professionals and volunteers dedicated to fire extinction, and also the civilian population (between 2001 and 2010, 24 dead and 191 wounded were reported in Spain [178]). Although firefighting operations centres have a wealth of information about the evolution of the fire, obtained through different sources (such as geographical maps, vegetation maps, weather, fire aerial and satellite images and simulation-based prediction), often such information is not properly transmitted through the command chain to the individuals on the field. Therefore, firefighters and civilians are sometimes surprised by the speed of spread of the fire and are trapped by it, get disoriented, and have no knowledge about which way to flee. To address such situations, the research group on High Performance Networking and Architectures from the University of Castilla-La Mancha proposed the development of the EIDOS system (EquIpment Destined for Orientation and Safety). In this system, a wireless sensor network (WSN) is deployed from the air once the fire is detected, to collect data about the fire behaviour in real time. For this, the WSN nodes are equipped with temperature, smoke or pressure sensors capable of detecting the presence of the fire [5]. The information obtained by the WSN is then transmitted directly to the mobile devices carried by the firefighters, without requiring any additional infrastructure. This PhD project is part of the EIDOS system. In particular, this research develops the tasks related to the development and distribution of the fire model. 1.2 Objectives First, we propose to design and develop a simulation framework that allows us to evaluate the performance of the contributions described in this thesis. This simulation framework also allows acquiring a detailed knowledge 1.3 Organization of the Thesis 17 about the problem, and also enables identifying the restrictions of existing solutions. Our second objective is to analyse the behaviour of different information dissemination mechanisms when they are employed in physical phenomena monitoring tasks. There must be a comprehensive survey of existing algorithms, which allows selecting the most relevant algorithms applicable to our scenario. Then, these algorithms will be implemented and evaluated through simulation by analysing their efficiency to propagate multiple simultaneous events throughout the network, the resources consumed in the network devices, and the overhead introduced in the shared wireless medium. Specifically, we are interested in determining which state of the arte technique is more suitable for this scenario. The third objective is to develop several mathematical models for the representation of forest fires. These models will be evaluated through simulation, by studying their efficiency in obtaining a good approximation to the current shape of the fire. This research will try to develop models with a low computational complexity (to enable local execution at each network node) and reduce the amount of information used (reducing storage and communication requirements). 1.3 Organization of the Thesis The structure of this thesis is broken down into the following chapters directly related to the objectives outlined in the previous section: Chapter 2 introduces basic concepts about WSNs, including hardware and software components, simulation methodologies used in its development, and areas of application. In addition, an introduction to collaborative information processing in WSNs focused on the tasks of locating, tracking, and contour mapping is made. 18 PhD Summary Chapter 3 presents the EIDOS system architecture, detailing the various modules that compose it. Then the implementation of the applications running on the network nodes and the handheld mobile devices is described. Due to the difficulty of evaluating our proposals in a real environment, we proceeded to develop a simulation tool. This chapter concludes with a description of this tool. Chapter 4 classifies existing information dissemination algorithms for WSNs, ranging from simple flooding to more sophisticated protocols that use directional antennas. Next, an exhaustive study of the techniques considered more appropriate for the EIDOS system is made. Finally, we proceed to its benchmarking (including the methodology and results), selecting the most suitable diffusion mechanism. Chapter 5 describes three models proposed for the representation of wildfires, which represent the main contribution of this thesis. Each proposal is evaluated and tuned separately, and then a comprehensive comparison of all of them is performed. Finally, Chapter 6 presents the conclusions obtained from the study developed and outlines some future works that remain open. Furthermore, it also reviews the scientific contributions derived from this work. In the following sections, the contributions that have been made are enumerated, and the main conclusions that can be drawn from this study are presented. Then, the various lines of work that can be explored in the future will be described. 1.4 Contributions The main contributions made by this work are: 1.4 Contributions 19 1. A simulation tool which allows, among other things, the assessment of diffusion algorithms, sensor networks, and distributed algorithms to determine fire shapes. This tool facilitated the task of evaluating the selected algorithms using realistic wildfire scenarios. 2. A study of the behaviour of some representative diffusion techniques when employed for disseminating multiple events to all nodes in a dense sensor network, for the task of monitoring a physical quantity spreading over a particular area. These diffusion techniques were implemented and compared using our simulation platform. 3. Three models have been developed for the representation of forest fires by means of distributed WSNs. To validate these proposals, the realistic fire simulation scenarios (presented as the first contribution) were employed. 4. The simplest model developed, assumes a circular area around each node detecting fire. The resulting shape will be homogeneous or heterogeneous depending on the radius of the circle used, which may be constant or may vary depending on the density of the deployed nodes or the density of the nodes that detected the fire. 5. The second proposal, which we call convex hull-based model, exhibits significant advantages over the previous one (in particular, better use of node memory and communication resources), while maintaining a good approximation to the real fire. 6. The third model proposed is based on a collection of arbitrary polygons whose shape evolves according to the fire detection events reported from a distributed WSN. 7. Finally, to conclude this thesis, a comparative simulation-based evaluation was conducted. The experiments compared all the models proposed for the representation of forest fires, and also helped to tune each algorithm, obtaining the optimal conditions (density of nodes and threshold distance) for which the system represents the fire more accurately. 20 PhD Summary 1.5 Conclusions Concerning information dissemination, the technique employing retransmission delays based on the covered area has shown to be more effective in all the aspects evaluated (percentage of successfully disseminated information, number of duplicate messages, and collisions in the wireless medium). As the only negative point, our evaluation has shown that this technique has higher network resources consumption, however not to the extent of representing a major drawback. As for the fire model, the limiting factors in our application are the node memory and communication requirements, with other factors such as the complexity of the algorithm ran by each node being in a second plan. The circle-based model has very little computational requirements, but consumes a large amount of memory and communication resources. Using homogeneous forms for sparse networks produces the best results with circles of intermediate size. With heterogeneous forms, the approximation based on the node density provides the best performance, showing for all densities much more precise adjustment, reaching a quality close to 90%. The convex hull-based model can correctly estimate up to 72% of the forest fire. Furthermore, this model can significantly reduce storage requirements compared to the previous model. In the particular case of the approach based on a single convex hull, it requires a very low and practically constant amount of points stored in memory, regardless of the network density or size of the fire. However, the estimates obtained are of substantially lower quality. Finally, the polygon-based algorithm provides an accurate approximation to the real fire and does not consume many resources in the network and in the nodes. In sparse networks, this approach gets close to an approximation 1.6 Future work 21 quality of 90%. Also, for dense networks, we obtain a quality of 95%. As for memory requirements, they were reduced to half of the resources required by the previous model. From this study, it can be concluded that the algorithm based on a collection of arbitrary polygons reduces node memory consumption, without affecting significantly to the quality of the fire approximation. 1.6 Future work The following are aspects not covered completely by this work, and that are considered interesting to be explored in the future: ¿ This study has made a first approach to diffusion algorithms, where nodes send messages notifying the arrival of the fire. A possible improvement would be to use information pre-processing techniques, taking advantage of the processing capability of the sensors, in order to reduce the amount of information to propagate, and hence the load on the wireless medium. Among these techniques we can find data fusion, which consists in to query the data received from nearby nodes and get a single output (such as the mean). Another type of pre-processing techniques is the aggregation of information, which is to exploit the correlation between the data obtained by neighbour nodes, which can be in time or space and to use techniques of data compression or coding [84]. In this case, it comes to reducing the bits to be transmitted while the readings obtained by the sensors do not change. Among the pre-processing techniques are also data filtering, which is based on not transmitting the readings that do not meet predefined conditions. ¿ Assign a certain area to fire detections, based on the density of neighbours of the detector node, and incorporate them to the 22 PhD Summary approximation, using a collection of points with a regular polygon shape (e.g., a hexagon). ¿ Improve the current model based on arbitrary polygons, by dynamically adapting the distance threshold as a function of the network density. ¿ Develop approximation models that are not restricted to the positions of the network nodes, but allow the insertion of virtual points. These techniques could be based on Bezier curves or B-splines. Bezier curves allow the representation of any curved shape and avoid unnecessary mathematical calculations. Such curves are defined by a series of points where the curve passes and other points outside that define the turning points of the curve. A B-spline is a generalization of a Bezier curve, which can avoid some problems of the previous approximation. Some authors have used P-splines or splines with penalties, which are two-dimensional smoothing methods, for representing mortality trends in a geographical area using a smaller number of parameters than other smoothing techniques [19]. ¿ Extend the existing simulation framework with new features that allow studying the impact of errors in the fire model including position errors, measurement and packet loss. When nodes are deployed, they must calculate its position, hence the rationale for studying the impact of position errors. Then the nodes begin to take sensor readings. It is then necessary to calculate the impact of these errors. Finally, nodes detect the fire and send a packet to all others, and thus, packet loss should be examined in the same way. After evaluating the fire model with errors we can find acceptable ranges of the parameters and ensure that our approach is reliable and works well using such parameters. Furthermore, according to the impact of each type of error, future changes or improvements could be defined. 1.7 Scientific contributions 23 ¿ Deployment of a real prototype that integrates the systems we are developing, in order to evaluate these techniques and identify potential problems that do not appear in a simulated environment. In this sense, the group of Networks and High Performance Architectures (RAAP) is currently establishing a partnership with the company Pyro Fire Extinction [186]. 1.7 Scientific contributions The PhD candidate has conducted this research as a member of the UCLM Research team RAAP. Within this framework, the PhD candidate has been involved in several research projects (financed by public national and regional entities), in partnerships with other centres, conducting research and publishing the results in various national and international forums. 1.7.1 Contribution to R&D Projects Below are the projects in which the PhD student participates or has participated. ¿ MOTEGRID: Grid Architecture for the Distributed Processing of the Information Collected by Wireless Sensor Networks (Arquitectura grid para el procesamiento distribuido de información captada por redes inalámbricas de sensores) - Funding: Junta de Comunidades de Castilla¿La Mancha, Consejería de Educación y Ciencia, PII1C09-0101-9476 - Participating Entities: Universidad de Castilla-La Mancha, Universidad de Murcia y Universidad Complutense de Madrid. - Duration: from April de 2009 to December 2012 - Main Researcher: Mª Carmen Carrión Espinosa 24 PhD Summary ¿ Server architecture, applications and services. (Arquitectura de servidores, aplicaciones y servicios). - Funding: Ministerio de Educación y Ciencia, TIN2009-14475-C04-03 - Participating Entities: Universidad de Castilla-La Mancha - Duration: from January 2010 to December 2012 - Main Researcher: Francisco J. Quiles Flor ¿ PRIMARIS: Support for collaborative processing of environmental information in wireless sensor networks. (Soporte para el procesamiento colaborativo de información medioambiental en redes de sensores) - Funding: Junta de Comunidades de Castilla¿La Mancha, Consejería de Educación y Ciencia, PII2I09-0067-3628 - Participating Entities: Universidad de Castilla-La Mancha, Galileo Geosystems, S.L. - Duration: from April 2009 to March 2010 - Main Researcher: Aurelio Bermúdez Marín ¿ EIDOS: Equipment Destined to the Orientation and Safety of firefighters in wildfire fightting operations. (Equipamiento Informático Destinado a la Orientación y Seguridad del retén en operaciones de extinción de incendios forestales) - Funding: Junta de Comunidades de Castilla¿La Mancha, Consejería de Medio Ambiente y Desarrollo Rural, PREG-07-25; Amount: 82.836€ - Participating Entities: Universidad de Castilla-La Mancha - Duration: from December 2007 to December 2010 - Main Researcher: Rafael Casado González 1.7 Scientific contributions 25 ¿ Improving the performance, reliability and services offered by high-performance cluster architectures. (Mejora de las prestaciones, fiabilidad y servicios ofrecidos por los servidores de altas prestaciones basados en arquitectura cluster) - Funding: Ministerio de Educación y Ciencia, TIN2006-15516-C04-02; Amount: 320.650€ - Participating Entities: Universidad de Castilla-La Mancha, Universidad Politécnica de Valencia y Universidad de Murcia - Duration: from October 2006 to September 2009 - Main Researcher: Antonio Garrido del Solo 1.7.2 Collaboration with other Research Centres The PhD candidate performed was a visiting researcher at the CISTER / INESC-TEC Research Unit, of the Polytechnic of Porto, Portugal from May 2012 to August 2012. The work performed during the stay was carried out under the supervision of Mário Alves and the cosupervision of researchers Nuno Pereira and Stefano Tennina. The core of the work performed was on the design and of evaluation of new algorithms to represent wildfires. 1.7.3 Publications The scientific contributions consubstantiated in peer-reviewed international publications are listed below. 1. Title: WSN-Based Distributed Forest Fire Shape Detection using Arbitrary Polygons Authors: Serna Moreno, María Ángeles; Casado González, Rafael Bermúdez Marín, Aurelio; Pereira, Nuno and Tennina, Stefano 26 PhD Summary Journal: International Journal of Sensor Networks (under review) I.S.B.N.: 1748-1279 Abstract: Fires consume every year millions of hectares of forest around the world, and the trend clearly shows an increase in their dimension and severity. This work aims to demonstrate how cheap wireless sensors can build dense networks with the objective to aid firefighting management operations. In particular, the primary goal is to accurately reconstruct the shape of a forest fire evolving over an area of interest, where the wireless sensor network (WSN) is deployed to provide real-time data to firefighters on the field. A key enabler of this approach is the mathematical model supporting the fire shape representation, which must (i) approximate the fire accurately and (ii) consume a small amount of processing and memory resources.In this work, we propose the use of a model based on a collection of arbitrary polygons whose shapes evolve according to the fire detection data coming from the distributed WSN. This paper significantly extends previous work by tackling the non-trivial challenge of dealing with non-convex fire shapes and by performing extensive validation of the proposed work through realistic fire simulations. Numerical evaluations demonstrate that the new model fits the fire shapes very accurately, and is able to dynamically adapt to the evolution of the fire fronts. 1.7 Scientific contributions 27 2. Title: Circle-based Approximation to Forest Fires with Distributed Wireless Sensor Networks Authors: Serna Moreno, María Ángeles; Bermúdez Marín, Aurelio; Casado González, Rafael Conference: EEE Wireless Communications and Networking Conference (WCNC 2013). I.S.B.N.: 978-1-4673-5939-9 Date and place of conference: April 2013. Shanghai, China Abstract: Disaster management is one of the most promising application fields of wireless sensor networks. In this work we focus on forest fire fighting operations. The role of the sensor network deployed in this scenario will consist in obtaining a representation (or a model) of the fire fronts, and distribute it to the firefighters in the area, in such a way they are aware of the risks they are assuming at each time. To achieve it, sensor nodes are able to determine their position, to detect the presence of fire (by monitoring the evolution of the ambient temperature), and to distribute this information wirelessly through the network. In this way, each firefighter maintains a fire representation consisting of a list of circles centered at the received positions. In this paper, we propose and analyze three criteria to assign the size of these circles. Simulation results show that a dynamic approach based on network density perceived by a node detecting fire offers an accurate approximation to the forest fire, without introducing a significant overhead. 28 PhD Summary 3. Title: Hull-based Approximation to Forest Fires with Distributed Wireless Sensor Networks Authors: Serna Moreno, María Ángeles; Bermúdez Marín, Aurelio; Casado González, Rafael Conference: The 8th International Conference on Intelligent Sensors, Sensor Networks, and Information Processing (ISSNIP 2013). I.S.B.N.: 978-1-4673-5501-8 Date and place of conference: April 2013. Melbourne, Australia Abstract: Disaster management is one of the most promising application fields of wireless sensor networks. In this application, the role of the sensor network usually consists in obtaining a representation or a model for a physical phenomenon spreading through the area. In this work we focus on forest fire fighting operations, proposing a distributed way for approximating the shape of the fire. In the simplest scenario, assuming the existence of an efficient dissemination layer, each network node obtains a representation of the forest fire starting from the complete list of positions reached by the fire so far. However, this approach makes an intensive use of node and network resources. In order to reduce these requirements, we propose incorporating in-network aggregation techniques. In particular, in this paper we analyze the employ of a shape composed of several convex hulls to represent the forest fire in a more compact way. As we will see, the use of multiple hulls representing different burning areas offers an accurate approximation to the forest fire, without introducing a significant overhead. 1.7 Scientific contributions 29 4. Title: Range-free Localization for Large-scale Environmental Sensor Networks Authors: García Polo, Eva Mª; Serna Moreno, María Ángeles; Robles-Gómez, Antonio; Bermúdez Marín, Aurelio; Casado González, Rafael Journal: International Journal of Information and Computer Science (IJICS) I.S.B.N.: 2161-6450 Abstract: Most of the applications of sensor networks require sensors being aware of their position. Usually, this position is estimated by means of a distributed localization algorithm, which assumes that the position of some network nodes is known a priori. In many cases each node obtains the area where it resides by intersecting the coverage areas of the nodes that it hears. Using circles to model these coverage areas introduces a high computational complexity. For this reason, some authors have modelled these areas by means of squares. In this paper we propose the employment of hexagons, in order to reduce the additional inaccuracy introduced by that shape. 5. Title: A Convex Hull-based Approximation of Forest Fire Shape with Distributed Wireless Sensor Networks Authors: Serna Moreno, María Ángeles; Bermúdez Marín, Aurelio; Casado González, Rafael; Kulakowski, Pawel Conference: The 7th International Conference on Intelligent Sensors, Sensor Networks, and Information Processing (ISSNIP 2011) 30 PhD Summary Date and place of conference: December 2011. Adelaide, Australia I.S.B.N.: 978-1-4577-0673-8 Abstract: Monitoring of physical phenomena is one of the most promising application fields of wireless sensor networks. In this work we focus on obtaining the shape of a forest fire. In this kind of applications, the information sensed by network nodes is usually transmitted to a base station located at the border of the network, where it is finally processed. However, such an approach requires that a large amount of data is transmitted through the network. In this paper, we assume that network nodes are able to collaborate in order to obtain an approximation of the forest fire shape without any base station, in a completely distributed way. We propose and analyze two techniques for performing this approximation. The first one makes intensive use of resources, while the second model incorporates an aggregation technique, reducing significantly resource requirements. 6. Title: Demo Abstract: Hazard Control in Forest Fire Fighting operations Authors: Casado González, Rafael; Bermúdez Marín, Aurelio; García Polo, Eva Mª; Serna Moreno, María Ángeles; Robledo De La Torre, Antonio; Fernández Díaz, Isabel Conference: The European Conference on Wireless Sensor Networks (EWSN 2011) Date and place of conference: February 2011. Bonn, Germany 1.7 Scientific contributions 31 Abstract: The EIDOS system is a WSN-based support system proposed for reducing hazardous situations for people working in forest fire fighting operations. This paper describes a tool that demonstrates the functionality of the system. In particular, the behaviour of the applications running at sensor nodes and handheld devices is shown by means of user-friendly graphical interfaces, which will be available for the attendants to the demonstration session.