Cloud Enabled Emergency Navigation using Faster-than-real

Cloud Enabled Emergency Navigation
Using Faster-than-real-time Simulation
Huibo Bi and Erol Gelenbe
Intelligent Systems and Networks Group
Department of Electrical and Electronic Engineering
Imperial College London
Cloud Enabled Emergency Navigation Using Fasterthan-real-time Simulation
Outlines
 Overview
 Related Work
 System Framework
 Routing Algorithms
 Simulation Model
 Experimental Results
 Conclusions
Overview
Objective
Current emergency navigation approaches direct evacuees
to exits in a real-time manner and casualties caused by the
poor decisions are only apparent at the end of the evacuation process and cannot then be remedied.
Our research aims to evaluate evacuation routes through a
cloud-based simulator and generate new routes for
“simulated casualties”.
Graph model
As an example of our approach, consider:
 A Graph: Representing three story canary wharf
shopping mall.
 Vertices (Points of Interest): Positions with landmarks
where evacuees can easily identify their locations by
uploading snapshots to cloud servers.
 Edges: Physical paths for evacuees such as corridors.
Related work
Cloud enabled emergency response system
 Compared with wireless sensor network based counterparts, cloudbased emergency navigation systems offer advantages such as high
processing power, large storage and high interoperability.
 Use on-site sensors or portable devices to gather sensory data and
offload intensive computations to cloud servers.
 Create a delayed feedback loop between live sensory data and routing
decisions.
Faster-than-real-time simulators in emergency management
 Most research focuses on inferring the spreading of hazard based on
predictive models and live sensory data.
System Framework
System architecture
 User layer
• Deployed on handhold devices.
• Collect sensory data.
 Cloud infrastructure layer
• Deployed on cloud severs.
• Consist of a data interpretation
module and a navigation module.
• Data interpretation module extracts
landmarks from the uploaded photos
and matches them with pre-stored
images.
• Navigation module is simulator that
contains many inter-connected
servers.
Routing Algorithms
Components
 Cognitive Packet Network with time metric (CPNST)
• CPNST pursues safe routes with shortest time to the exit.
• We use CPNST as the basis algorithm to guide evacuees.
 Time-dependent Dijkstra’s algorithm
• A variant of Dijkstra’s shortest path algorithm.
• Replace the original distance metric with a time metric.
Procedures
1. Perform CPNST to generate routes for evacuees in the simulator;
2. Reassign routes for perished evacuees in the simulation;
3. Send the generated routes to evacuees.
Cognitive Packet Network
Cognitive Packet Network (CPN)
Intelligent capabilities for routing and flow control are concentrated in the
packets.
 Basis
Random Neural Networks
 Components
• Smart Packets (SPs)
Search paths and collect information with regard to pre-defined goal functions.
• Acknowledgements (ACKs)
Bring back the information collected by SPs.
• Dumb Packets (DPs)
Carry the payload. In the context of emergency evacuation, evacuees are
considered as DPs.
Variations for a cloud-based environment
 Each CPN node is deployed on a cloud server and SPs are used to
gather interested information from other servers.
 Each server is also associated with a position with landmarks.
Time-dependent Dijkstra’s algorithm
Time-dependent metric
e
T ( ( i) ,  ( i  1) ) 
E ( ( i) ,  ( i  1) )
Vs
q
 tc ( (i))
Term 𝐸 𝑒 (𝜋 𝑖 , 𝜋 𝑖 + 1 ) is the effective length of the edge between node
𝜋 𝑖 and 𝜋 𝑖 + 1 .
Term 𝑉𝑆 represents the average speed of civilians.
𝑞
Term 𝑡𝑐 (𝜋 𝑖 ) denotes the queueing time after an evacuee reaches node
𝜋 𝑖 .
Queueing time at a node
q
t c ( ( i) ) 
N
d
c
q
 (i)
Term 𝑁𝑞𝑐 is the number of queued civilians when the evacuee arrives and
𝑑𝜋(𝑖) is the departure rate of node 𝜋(𝑖).
Simulation model
DBES – Existing Tool in Prof.
Gelenbe’s Group.
 Agent based.
 Graph based.
Assumptions
 A data center with massive
servers is used for decision
support.
 Evacuees can communicate with
the cloud over 3G or WI-FI.
 Fire starts near a main channel.
 Evacuees are randomly scattered in the building.
 Cloud servers can obtain the position of individuals by matching the image
snapshots uploaded by evacuees.
•
•
Results: Average Percentage of Survivors
CPN with time metric (CPNST) performs better than Dijkstra’s algorithm in higher
occupancy rates due to its embedded congestion-ease mechanisms.
The proposed simulation-based algorithm achieves higher survival rate than CPNST.
This is because CPNST may take the risk to traverse potential hazard areas in order to
reduce the evacuation time and our proposed algorithm can generate new paths for
perished evacuees in the simulation.
Results: Number of two-way information exchanges
•
•
One two-way information exchange is defined as the process of an evacuee
uploading a snapshot and gaining the suggested path from the Cloud.
As expected, our proposed algorithm achieves the lowest number of data
exchanges with cloud servers because can send final paths to evacuees when
a disaster breaks out rather than making periodically decisions based on live
sensory data.
Conclusions
Conclusions
 The proposed simulation-based routing algorithm can improve the survival
rate of an evacuation process.
 Owing to the hazard predictive model, the proposed algorithm can
calculate desired paths only based on the initial fire location and
distribution of evacuees. Hence, the information exchanges between
evacuees and cloud severs is reduced.
Cloud Enabled Emergency Navigation Using Faster-thanreal-time Simulation
Thank you!