Towards Adaptive Sensor Data Management for Distributed Fire

0 downloads 0 Views 413KB Size Report
Abstract— We introduce a novel strategy for data processing in. Wireless Sensor Networks (WSNs) in the case of emergency fire evacuation with stringent delay ...
Eleventh International Conference on Mobile Data Management

Towards Adaptive Sensor Data Management for Distributed Fire Evacuation Infrastructure Andrii Cherniak, Vladimir Zadorozhny School of Information Sciences University of Pittsburgh Pittsburgh, PA, USA [email protected], [email protected] in real time. While e-WSNs are intended to handle large concurrent streams of data, the sensornet performance dramatically degrades with increases in network size and data rate [Sha09, Xu04]. One way to improve utility of large e-WSNs is to tune up the information flow in such a way that the maximal amount of data is delivered with minimal resource consumption. Achieving this objective is not always possible. The factors impacting successful information delivery in an e-WSN are numerous (e.g., link quality degradation, congestion, no route availability, packet collisions, etc.). Their combined effect is hard to measure for different network configurations and application scenarios. While many optimization techniques have been proposed for WSNs (see review of related works in Section V), complex interplay of all factors impacting the performance makes it difficult to explicitly tune up and optimize the data delivery in large sensornets. In this paper, we introduce a light-weight adaptive strategy to improve the functionality of e-WSNs by minimizing the amount of delivered sensor data without compromising the application requirements. Instead of considering the WSN as a black box and devising a generic optimization strategy based on constraints and data losses, we propose to treat a sensornet as a complex adaptive system (CAS) [Mil07]. In this case, the complex task of emergency fire management can be reduced to the task of localized decision making by individual sensors, based on locally-observable conditions. We will demonstrate that such decisions converge into a desirable information delivery pattern. A most notable characteristic of our approach is its tolerance for considerable information losses while meeting stringent application constraints. The rest of the paper is organized as follows. Section II introduces an e-WSN system model for emergency evacuation. Section III elaborates on our proposed e-WSN evacuation strategies. Section IV reports on experimental analysis and demonstrates the efficiency of our approach. Section V considers related works, followed by our conclusions in Section VI.

Abstract— We introduce a novel strategy for data processing in Wireless Sensor Networks (WSNs) in the case of emergency fire evacuation with stringent delay constraints. Such networks should perform distributed emergency assessment, continuous emergency monitoring, and dynamic selection of optimal evacuation strategies. The obvious complexity of these tasks restricts applying existing WSN optimization solutions. Our approach is based on considering the WSN as a complex adaptive system where a complicated task of multi-agent scheduling is factored into a set of smaller subtasks. In this case, decisions made locally by individual sensors can efficiently converge into desirable information processing patterns. A notable feature of our method is its scalability, which allows the sensornet to operate with sufficient quality of service under heavy information loads. We demonstrate the utility of our approach using different fire evacuation scenarios. Keywords-wireless sensor networks, emergency evacuation, complex adaptive systems, emergent behavior

I.

INTRODUCTION

It has become increasingly popular to consider Wireless Sensor Networks (WSNs) as an implementation platform for mission-critical applications. A prime example of such an application is emergency fire evacuation, which involves timely emergency assessment, continuous environment monitoring and selection of an optimal evacuation strategy. Numerous large-scale disasters (e.g., the King’s Cross underground station fire in 1987 and the Mont Blanc tunnel fire disaster in 1999) demonstrated severe deficiencies in the emergency evacuation systems. Such disasters follow similar patterns, including the presence of many people in enclosed spaces, limited possibilities of escape, and lack of awareness of the situation. Fire alarms or indirect warning signs such as the presence of smoke, communicate very little about the location of the source of fire, its severity, proper evacuation routes, and the amount of time that each of the evacuation routes will remain available. Lack of this vital information commonly results in wasting of time, trial and error behavior, panic, and, as a consequence, higher casualties. The application requirements for sensor data freshness, completeness and accuracy are especially high for fire evacuation systems, as human lives are often at stake. At the same time, the well-known resource constraints have a very strong impact on the emergency fire management WSN (eWSN), in which nodes sense and exchange critical information 978-0-7695-4048-1/10 $26.00 © 2010 IEEE DOI 10.1109/MDM.2010.53

II.

SYSTEM MODEL

Each sensor in an e-WSN maintains a system model that supports decision-making for emergency evacuation. According to this model, the evacuation space is split into square cells of equal size. In a real situation, those cells can correspond to actual locations in a building, or zones of 151

specifically, we define data cost as a number of messages normalized by a maximal number of exchanges under given circumstances. N is to maintain the highest A major objective of e-WSN possible values of NR and LT wh hile minimizing data cost. Next, we will introduce our adapttive approach to solve this problem. As a comparison baselinee, we also consider a nonadaptive technique called informatio on diffusion.

responsibility controlled by sensors. This sppace is populated with people, and there is at least one source of fire and at least one exit. Every cell S contains a wireless sensor that monitors conditions and controls evacuation from that ccell. Every sensor maintains the information required to m make evacuation decisions. Part of this information iss received via communications from other sensors in thhe e-WSN. This information includes the following items: • Hazard_time – estimated time before fire reaches cell S. • Delay – estimated time which a persoon needs to leave the building, starting from cell S, or to reach another specific point inside the building. • Time_to_ignite - time required for thhe cell S to catch on fire, assuming that neighboring cellls are on fire. • Waiting – estimated time a person speends in the cell. • Occupancy – an estimated maxim mum number of people which the cell S can accommoddate. • Throughput – an estimated number off people who will be allowed to move from cell S to oother cells within one time unit. Sensors assume that each person hass a certain life expectancy and that moving in the proximity oof fire reduces it. A major objective of the e-WSN is to avooid reducing life expectancy as much as possible while evacuatting the maximal number of people. Apparently, there is a veryy sensitive tradeoff in reaching this objective that the e-WSN N should carefully evaluate. Consider Figure 1, which shows a simulated fire evacuation scenario in a building with one soource of fire, one exit, 441 cells, and 2205 people to be evvacuated. At the beginning, all people go towards the exit, whhich is located in the middle of the top border of the space. At time tick 33, fire reaches the exit. A snapshot of the evacuationss when the exit is on fire is shown to the right of Figure 1. Red cells correspond At that point, the to fire and white cells correspond to people. A only available evacuation option is to go awayy from fire to the safest place. After time tick 33, the number oof rescued people remains constant and the number of victims inccreases.

III.

USING WSN FOR EMER RGENCY EVACUATION

A. Non-adaptive information diffussion strategy Information diffusion (InD) sttrategy tries to lay out the fastest route to the exit, where the im mpact of the fire is below a given threshold. This method assum mes that information about the whole route -- or a significant po ortion of it -- is available to a sensor for decision-making. This information is a part of the global knowledge that is maintaineed by the e-WSN. The InD method uses a threat function, whicch shows how proximity to the fire impacts life expectancy. It I is obvious that keeping complete information about each po ossible evacuation route for each sensor in an e-WSN does no ot scale. This is why InD maintains the information about n points, p which have the most significant impact on life expectanccy over a chosen route – socalled critical points. If possible, sen nsors select the fastest route to the exit with highest acceptable life expectancy. Otherwise, if a safe route to exit does not exist, people are navigated to the cells with the longest hazard_time. ure 2, sensor 3 detects two In the example, shown in Figu routes to the exit: 3-2-1 and 3-6-5-4 4-1 (numbers denote critical points). If both routes are safe, then 3-2-1 will be selected. If all routes are not safe, it suggests a diirection away from the fire (towards the sensor with the longest hazard_time).

Figure 2. Critical points p

B. Adaptive State/Action Strategy In contrast to the information n diffusion approach, the adaptive method does not require an ny global information: all of the actions are taken based on the lo ocal state of a sensor and its direct neighbors. The network iss partitioned into a finite number of states as shown in Figu ure 3. Here HT stands for hazard_time threshold, and DT for distance_to_exit threshold. A sensor takes actions according to its current state.

Figure 1. Fire evacuation dynam mics

We use several metrics to estimate thee performance of the e-WSN. They include normalized rescuinng (NR), life time (LT), and data cost (DC). Normalized rescuingg shows the ratio of the number of successfully-evacuated people to the maximum number of people who could havee been evacuated before the fire reaches the exit. Life time is the ratio of how m possible time, long a person stayed alive to the maximum which is the hazard_time of the last cell caughht on fire. Finally, data cost shows how much information was exchanged between sensors in the e-WSN during the evaccuation. A higher data cost increases data processing overhead, nnetwork load and congestion, as well as depleting the sensor’ss battery. More

State S1 S2 S3 S4

Definition n 0

, , 0,

S5

Figure 3. State deefinitions

152

0 0

area with 21*21 cells and 2205 people inside. We set one source of fire in the middle of the left edge, and placed the exit is in the middle of the top edge. All measurements were averaged over 10 different ignition time setups per fire pattern. Time_to_ignite parameters were initialized with values taken from the log-normal distribution , . We simulated the results for three fire patterns: deterministic ( 1, 0 ), moderately stochastic ( 1, 2 ), and highly stochastic ( 1, 3 ). We simulated the environment for different sizes of areas of S1 and S3 (HT and DT values). We observed a few expected tendencies in rescuing performance: it increases with the increase of DT(size of S3), and decreases with the increase of HT(size of S1). With the increase of the area of S3, more people were aggressively directed towards the exit. With the increase of the area of S1, more people were navigated away from both the fire and the exit area, thus reducing the number of people in S3 and decreasing rescuing performance. The increase of the area of S1 increased the life time, because the sensors direct people away from the fire; thus, the fire has little influence on human lives. In addition, for highly-stochastic fire, we observe that simultaneously high values of HT and DT lead to the improvement in rescuing performance. This happens because the fire front is not circular for a highly stochastic fire, and directing people away from the fire actually results in partially propelling people towards the exit area, which enhances the rescuing performance. In general, the state-action algorithm shows similar or improved rescuing and life time performances as compared to information diffusion. However, it requires about 30% more messages. This difference comes as the result of the information spread criteria.

Assuming that fire tends to behave stochastically with spikes in spreading, a good strategy for the victims in the vicinity of the fire is to go further from the fire and then head towards the exit. Such behavior would be desirable in S1. Since the density of people near the exit is expected to be high, any overreacting to fire dynamics in S3 would severely reduce the evacuation performance. A good strategy here would be to direct people straight to the exit. If sensors are far from the fire and the exit (S2), then the sensors should make their decisions based on the estimated distance to the exit and to the fire. During the evacuation process, sensors may change their states. As the fire spreads, sensors from S1 will be in the fire (S4); a group of sensors from S2, which are close to S1-S2 border, will change their states to S1. Similarly, as the fire moves closer to the exit, sensors from S3 will move to S4. However, because of the environmental constraints, not every direct state-to-state transition is possible. For example, the system can only transit from S2 to S5 via S1. Every sensor uses local information to assess its state and to perform a set of actions. In general, the primary goal of a successful evacuation is to take action in order to simultaneously increase hazard_time and decrease delay on the way to the exit. In many cases, it is not possible to choose a neighboring cell which satisfies both conditions. This is why each sensor uses Pareto-optimal criteria to weight both the change in hazard_time and delay parameters while choosing the neighboring cell to direct the evacuation. Each sensor computes the utility of its direct neighbors, which is the weighted sum of the differences in hazard_time and delay. The sensor will direct people towards the neighbor with the highest utility. All these operations require only locally-available knowledge, and are computationally inexpensive. Considering different tolerances to suboptimal decisions in different states, we hypothesize that S1 and S3 are less sensitive to outdated information about the environment then S2. Thus, we could reduce frequency of data updates in S2 without significant performance degradation. We will test this hypothesis in Section IV.

A. Reduced information updates Here we come back to our previously-stated assumption that S1 and S3 should be sensitive to the frequency of updates, while S2 should demonstrate weaker dependency. The results are shown in Figure 1. For a deterministic fire, rescuing performance does not change with a decrease in the frequency of updates in S2 only. However, it decreases when the interval between updates increases for all states: S1, S2 and S3. It demonstrates that even if rare updates make people move via suboptimal routes from S2 to S3, there are still too many people near the exit area. Frequent updates in S3 help to navigate more people to the exit, regardless of the supply from S2. The situation changes when the fire is moderatelystochastic and highly-stochastic. When the update frequency decreases for S2 only and for all states: S1, S2, and S3, rescuing performance increases till it reaches its maximum for update interval =3 ticks. Then rescuing performance for reduced updates in S2 only remains at the same level (for moderately stochastic fire) or decreases (for highly stochastic fire). For reduced updates in all states, with further increase in the update interval decreases rescuing performance. However, rescuing performance for reduced updates in S2 only shows same or better values, than rescuing performance for reduced updates in all states. This confirms our hypothesis about the lack of tolerance to suboptimal decisions in the areas with the highest density of people. The first conclusion is that the algorithm should react quickly to the changing environment. The second

C. Parameters Estimation and Data Dissemination Both InD and State/Action require the knowledge of hazard_time and delay to perform route calculation. To estimate fire propagation parameters, we implemented a distributed token-based method, which can be considered as an instance of heterogeneous team learning [PL05]. The main idea of this approach is to store and process fire information in a thin layer of sensors which surrounds the fire -- a processing layer (PL). Sensors assign themselves to this layer dynamically as soon as they detect the fire in the vicinity. Every sensor in the processing layer stores a limited number of the most recent measurements (history) in a buffer. If fire captures a PL sensor with a history, the sensor sends its entire buffer to the closest sensor in the processing layer. IV.

EXPERIMENTAL RESULTS AND ANALYSIS

We performed experiments to compare the performance of both InD and State/Action methods, by using Netlogo (simulator for complex adaptive systems) [Net1]. The evacuation space was set as in (Figure 1, left): it was a square

153

Deterministic fire

Moderately stochastic fire

Highly stochastic fire

NR

LT

DC

Fig 4. Reduced information updates

conclusion is that frequent updates not always lead to the best rescuing performance. We will explain this phenomenon in the next subsection. For a deterministic fire, life time performance achieves its local maximum for update = 3 ticks for reduced updates in S2 only. For update reduction in all states (S1, S2 and S3), it sharply goes down even for a slight decrease in the frequency of updates. The same tendency is observed for moderately and highly stochastic fires. Life time performance records for update reduction in all three states (S1, S2, S3) are significantly lower the corresponding records for update reductions for state S2 only. One of the reasons for that is inaccuracy in the hazard_time calculation. If a cell has n neighbors on fire, the resulting time for this cell to become on fire will be approximately n times less than the average ignition time. That is why frequent updates help to adjust the difference between what was assumed and what is actually happening in the system. The second reason is the stochastic behavior of fire. If a fire spike happens while there are no updates in the system, the information about hazard time becomes obsolete. Thus, there are sensors which still believe that they should remain in S2; however, they should have changed their state to S1 and adjusted their strategy for decision-making. For reduced information updates in S2 only, data cost falls below what the information diffusion strategy has produced for either deterministic or moderately stochastic fires. Meanwhile, rescuing and life time performances are maintained at the same or better level. This is not true for a highly stochastic fire. Here, the total number of message exchanges is slightly higher than for information diffusion, and life time has a 0.5 % higher

value, which could be treated as roughly equal. Information cost for reduced updates in all three states is significantly below the costs for information diffusion and updates in S2 only. The explanation for this result lies in the communication schema. When the update interval is large enough, the only information being sent in the reduced number of updates in states S1, S2, and S3, are the measurements for the average speed calculation in the fire front area. If updates are reduced in S2 only, then updates in S1 and S3 are performed at every time tick, which requires sending more data. B. Emergent behavior and multi-factor systems We have observed that the global human behavior during evacuation emerges from the local patterns. When the route to the exit is wide, the interactions between humans are minimal; thus, they do not obscure the route from each other. When the route is narrow, then people have to compete to transverse it. This is the mechanism of micro-crowd formation. Frequent updates about current cell delays perform a sort of load balancing by sending people via a wider route, and mitigating the effect of local crowds, as shown in Figure 5. Thus, frequency of updates changes the speed of human flow in the system. The change in flow speed results in the global effect of producer-consumer interaction. The “Producer” is S2, because it accommodates the largest number of people. The “Consumer” is the exit, which allows people to leave the system. With faster updates, people reach the exit area more quickly, thus increasing exit performance. When the exit disappears (gets in fire), people move quickly from the former

154

exit towards the safest place. When updates arrive less frequently, people reach the exit area at a slower pace. However, humans will behave in a more stubborn manner: even if the exit disappears, they will continue to try to reach the exit area for a certain period of time. We will apply this concept of emergent patterns to explain the dynamics of plots for reduced information updates in S2 only, shown in Figure 6.

network aggregation has also been proposed to save energy by reducing the amount of communications at the expense of extra computation [Mad07, You02]. TAG [Mad07] and Cougar [Yao02] generate query routing trees similar to relevant work in DTA [Zad04]. TiNA [Sha03] is a middleware layer sitting on top of either TAG or Cougar. TiNA employs query semantics (in particular, Quality of Data) and can reduce energy consumption significantly by eliminating redundant transmissions. Prior work in the area of fire evacuation is considerable. The following references are those closely related to the current work. The FIREGRID project [Ber05] and [Lim07] use real-time data on hazard spread, obtained from a wireless sensor network (WSN), to generate an emergency response scenario. One important weakness of these methods is centralization: data is being processed on a computer directly connected to the WSN. In case of network partitioning, the centralized approach will not be able to do its job. The next group of algorithms ([Tse06], [Pan06] [Gos04], [Tsu09], [Bar07] and [Cr06]) establish the core of distributed algorithms for WSNs to assist in disaster evacuation. They delineate safe routes towards exits, taking into consideration actual fire spread. In addition, [Bar07] proposes to predict the dynamics of fire by gathering information from detectors and substituting measured data into the fire spread model. On the contrary, [Tsu09] proposes to monitor the behavior of people to

Figure 5. Micro-crowd formation, a) outdated information b) frequent updates

For a deterministic fire, life time achieves its maximum value at update interval = 3 ticks, and then monotonically decreases. With an update in every tick, people reach the exit and leave through it at the quickest pace. When updates happen every 3 ticks, people reach the exit more slowly; however, when the exit disappears, the exit area is less crowded than in the case of updates in every tick. This creates a faster rerouting procedure for people in the exit area to travel to the safest part of the building, as compared to the case where updates occur in every time tick. When updates occur every 40 People near exit

Deterministic fire

Moderate stochastic fire

Highly stochastic fire

Figure 6. Agent dynamics for reduced updates in S2 only (UI – update interval)

ticks, the speed of people becomes even lower, but sensors are more persistent in their decision strategy. Sensors keep sending people to the exit area long after the exit disappeared. The exit area becomes overcrowded after the exit disappears, which reduces the life time performance. The same pattern is observed for a moderately stochastic fire. At the time when the exit disappears (about 60 ticks), algorithms with update intervals of either 3 or 40 ticks keep sending people to the exit area. Finally, when fire is highly-stochastic, we do not have a clear pattern of dynamics near the exit. The authors assert that, in this case, a person’s natural impulse to go to the exit and stay away from the fire would prove to be more effective than directions from the sensornet.

map safe evacuation routes. The aforementioned algorithms provide a strategy for evacuation from a building, which is only valid for a single person. [Lu03] was the first attempt to construct a bridge between computationally-expensive optimal methods and naive heuristics by introducing capacity constraints on the routes. It aimed to efficiently generate evacuation scenarios which consider the mutual influence of people. At its heart is the principle of deterministic evacuation: people who are closer to the exit will leave the building earlier; thus, they will introduce delays which other people will face. The main argument against this approach is determinism in behavior: usually people behave non-deterministically, which may be critical at some point and which will require constant route recalculation. The second important point is the absence of fire or any other source of hazard in the model: it does not predict how the optimal schedule should be modified in the presence of a threat. And finally, to the best of our knowledge, none of these algorithms address the issue of having no exits available for

V. RELATED WORK Researchers have been developing intelligent cost-based strategies for optimizing the data delivery in sensor networks [Zhe03, Sch02, Ye02, Che01]. In [ZK03], the authors proposed a cross-layer design for power management that utilized knowledge about the route set-up and packet forwarding. In-

155

evacuation: every method assumes that people will be able to leave the building. Also proposed are data dissemination schemes such as SPIN [HKB] which uses flooding; gradient-based Directed Diffusion [Est99]; clustering-based LEACH [Hei00], and GAF [Xu01]. Wave scheduling [Tri04] minimizes packet collisions by carefully scheduling the sensor nodes. It results in energy savings at the expense of increased message latency. Synopsis Diffusion [Nat04] proposes a multipath routing scheme, which is more robust than tree-based TAG and avoids message double-counting. VI.

[Mad07] S. Madden, M. Franklin, J. Hellerstein, W. Hong, “TAG: a Tiny AGgregation service for Ad-Hoc sensor networks,” Proc. of the 5th symposium on Operating systems design and implementation, December 09-11, 2002, Boston, Massachusetts [Mil07] J. Miller, S. Page, “Complex Adaptive Systems,” Princeton University Press, 2007. [Net1] NetLogo homepage: http://ccl.northwestern.edu/netlogo/ [Nat04] S. Nath, P. Gibbons, S. Seshan, Z. Anderson, “Synopsis diffusion for robust aggregation in sensor networks,” Proc. of the 2nd international conference on Embedded networked sensor systems, November 03-05, 2004, Baltimore, MD, USA. [Pan06] M. Pan,C. Tsai, Y. Tseng, “Emergency guiding and monitoring applications in indoor 3D environments by wireless sensor networks,” International Journal of Sensor Networks, vol.1, no. ½, pp. 2-10, 2006 [Sha03] M. Sharaf, J. Beaver, A. Labrinidis, P. Chrysanthis, “TiNA: a scheme for temporal coherency-aware in-network aggregation,” In MobiDE, 2003 [Sch02] C. Schurgers, V. Tsiatsis, M. Srivastava, “STEM: Topology Management for Energy Efficient Sensor Network,” Prov. of IEEE Aerospace Conf., 2002. [Sha09] D. Sharma, V. Zadorozhny, “Adaptive Information Delivery in Data-Intensive Sensor Networks,” Technical Report, University of Pittsburgh, 2009. [Tsu09] T. Tsunemine, E. Kadokawa, Y. Ueda, J. Fukumoto, T. Wada, K. Ohtsuki, H. Okada, “Emergency UrgentCommunications for Searching Evacuation Route in a Local Disaster,” IEEE Consumer Communications and Networking Conference 2008, pp.1196-1200, Jan. 2008. [Tse06] Y. Tseng, M. Pan, Y. Tsai, "Wireless Sensor Networks for Emergency Navigation," IEEE Computer, vol. 39, no. 7, pp. 55-62, July 2006 [Tri04] N. Trigoni, Y. Yao, A. Demers, J. Gehrke, R. Rajaraman, “WaveScheduling: energy-efficient data dissemination for sensor networks, “ In Proceedings of the 1st international workshop on Data management for sensor networks: in conjunction with VLDB 2004, August 30-30, 2004, Toronto, Canada. [Xu01] Y. Xu, J. Heidemann, D. Estrin, “Geography-informed energy conservation for Ad Hoc routing,” Proc. of ACM MobiCom '01, Rome, Italy, July 2001 [Xu04] N. Xu, S. Rangwala, K. Chintalapudi, D. Ganesan, A. Broad, R Govindan, D Estrin, “Wireless sensor network for structural monitoring,” Proc. of the 2nd ACM Conference on SENSYS, 2004 [Yao02] Y. Yao, J. Gehrke, “The Cougar approach to in-network query processing in sensor networks,” ACM SIGMOD Record, v.31 n.3, September 2002 [You02] M. Younis, M. Youssef, K. Arisha, “Energy-Aware Routing in Cluster-Based Sensor Networks,” Proc. of the 10th IEEE International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunications Systems (MASCOTS'02), p.129, October 11-16, 2002. [Zad04] V. Zadorozhny, P. Chrysanthis, P. Krishnamurthy, “A framework for extending the synergy between MAC layer and query optimization in sensor networks,” Proc. of the 1st international workshop on Data management for sensor networks: in conjunction with VLDB 2004, August 30-30, 2004, Toronto, Canada. [Zhe03] R. Zheng, J. Hou, L. Sha, “Asynchronous wakeup for ad hoc networks,” Proc. of the 4th ACM international Symposium on Mobile Ad Hoc Networking &Amp; Computing (Annapolis, Maryland, USA, June 01 - 03, 2003). MobiHoc '03. ACM, New York, NY, 35-45. [ZK03] R. Zheng, R. Kravets, “On-demand Power Management for Ad-Hoc Networks,” Proc. of IEEE INFOCOM Conf., 2003. [Ye02] W. Ye, J. Heidemann, D. Estrin, “An Energy-Efficient MAC Protocol for Wireless Sensor Networks,” Proc. of IEEE INFOCOM, 2002.

CONCLUSION

We introduced a scalable adaptive approach improving the utility of WSNs for fire evacuation. Our approach considers the sensornet as a complex adaptive system that performs emergency fire management based on localized decisions made by individual sensors. Thus the complex task of multi-agent scheduling can be factored into a set of more simple subtasks. This approach considerably reduces the amount of delivered sensor data without compromising the stringent application requirements. REFERENCES [Bar07] M. Barnes, H. Leather, DK. Arvind, “Emergency Evacuation using Wireless Sensor Networks,” Proc. of the 32nd IEEE Conference on Local Computer Networks (LCN’07), p. 851-857, Oct 2007 [Ber05] D. Berry, A. Usmani, J. Torero, A. Tate, S. McLaughlin, S. Potter, R. Baxter, M. Bull, M. Atkison, “FireGrid: Integrated emergency response and safety engineering for the future built environment,” Proc. UK e-Science Programme All Hands Meeting (AHM 2005), 19-22 September, Nottingham, UK [Che01] B. Chen, K. Jamieson , H. Balakrishnan, R. Morris, “Span: An energyefficient coordination algorithm for topology maintenance,” Proc. of the 7th annual international conference on Mobile computing and networking, p.85-96, July 2001, Rome, Italy. [Cr06] Christakos, C. “Sensor Networks Applied to the Problem of Building Evacuation: An Evaluation in Simulation,” Proc. of the 15th Annual IST Mobile and Wireless Conference. Mykonos, Greece, June 2006 [Est99] D. Estrin, R. Govindan, J. Heidemann, S. Kumar, “Next century challenges: scalable coordination in sensor networks,” Proc. of the 5th annual ACM/IEEE international conference on Mobile computing and networking, p.263-270, August 15-19, 1999, Seattle, Washington, United States. [Gos04] M. Gosalia, K. Lin, A. Redfern, S. Romanovsky, N. Shah, D. Steingart, S. Teh, N. Turner, W. Watts, X. Yang, P. Levis, “Smoke: Mote Powered Fire Evacuation,” http://cents.cs.berkeley.edu/retreats/winter 2005/nest-1-05-re.ppt, 2004 [Hei00] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, “Energy-Efficient Communication Protocol for Wireless Microsensor Networks,” Proc. of the 33rd Hawaii international Conference on System Sciences-Volume 8 - Volume 8 (January 04 - 07, 2000). HICSS. IEEE Computer Society, Washington, DC, 8020. [Lu03] Q. Lu, Y. Huang, S. Shekhar, “Evacuation Planning: A Capacity Constrained Routing Approach”, Proc. of the First NSF/NIJ Symp. on Intelligence and Security Informatics, pp. 111-125, June 2003 [Lim07] Y. Lim, S. Lim, J. Choi, S. Cho, C. Kim, Y. Lee, "A Fire Detection and Rescue Support Framework with Wireless Sensor Networks," Proc. of International Conference on Convergence Information Technology, 2007, pp. 135-138

156

Suggest Documents