An Intelligent and Adaptable GridBased Flood Monitoring and Warning System

1
An Intelligent and Adaptable Grid-Based Flood
Monitoring and Warning System

• Phil Greenwood

2
The Problem

• Flooding is becoming a more common occurrence
• Climate change
• Land use
• Cost of damage correlates with
• Rate of flow
• Depth of water
• Warning time given
• To cope with this problem initiatives taken to
• Improve flood defences
• Raise public awareness
• Improve flood warning systems

3
Traditional Approaches

• Deploy sensors at flood prone sites
• Collect data manually or transmitted using GSM
  technology
• Data then processed using spatial or point-based
  prediction algorithms
• The results from these algorithms can be used to
  issue flood warnings

4
Limitations

• Rigid separation between on-site sensor network
  and off-site computation Grid
• Tends to be bottleneck
• The sensors used are computationally dumb
• They simply record and store/transmit data
• Data holds valuable information on how the
  sensors should behave
• No variation in the sensor behaviour possible
• Turn device off when un-interesting events
  occurring
• Increase frequency of measurements made
• No dissemination of warnings

5
Proposed Approach

• Increase local computational power of sensors
• Allow the local execution of flood prediction
  algorithms i.e. light-weight Grid
• Adaptation of the wireless sensor network
• Support a wider range of hardware
• Novel techniques for flood prediction and
  analysis
• Timely distribution of flood warnings
• Proactive and passive warnings
• SMS/Audio-Visual/Web

6
The GridStix Platform

• Consists of a variety of hardware and software
• Gumstix hardware platform
• Lancasters GridKit middleware platform
• Various networking technologies
• Flood prediction algorithms

7
Gumstix (1)
8
Gumstix (2)

• Specs
• 400Mhz Intel XScale processor
• 64Mb RAM
• 16Mb Flash Memory
• Variety of I/O Mechanism
• Standard Ethernet port
• Compact Flash slot
• Storage
• 802.11b Networking
• GPRS
• GPIO Lines for sensor connectivity
• On-board Bluetooth Radio

9
Power Consumption

• Significantly higher power consumption than
  devices used in traditional sensor networks
• Berkley Motes typically use 54mW
• Gumstix use 1W
• Can be powered using a combination of batteries
  and solar panels
• One 15cm2 solar panel output of 1.9W
• 6v 10AH battery
• Aggressive power management

10
GridKit (1)

• Provides key functionality to implement Grid
  behaviour
• Service Binding
• Resource Discovery
• Resource Management
• Security
• Based on the OpenCOM component model
• Rich support
• Stripped-down deployments
• Overlay support
• Used to implement networking service not provided
  by the underlying network type

11
Overlay (1)
12
Overlay (2)
13
Supported Adaptations

• CPU Adaptation
• Throttle CPU frequency
• Overlay Adaptations
• Swap overlay components to alter topology
• Physical Network Adaptations
• Switch network types

14
Adaptation Scenario 1

• Changes in Criticality
• Need to conserve power in normal operating
  conditions
• Operate at lowest CPU frequency
• Poll sensor infrequently
• Potential Flooding Detected
• GridStix can increase CPU frequency to execute
  prediction algorithms quicker
• Data can also be collected more frequently to
  improve the accuracy of the predictions

15
Adaptation Scenario 2

• Adapting to Node Failure
• Need to increase the robustness of network when
  flooding is predicted
• Do this without changing network type
• Switch overlay types
• Shortest path trees consume less power
• Fewest hop trees are more robust

16
Adaptation Scenario 2 cont.
Fewest Hops
Shortest Path
Root
Edge x
Edge x
Node B
Node C
Trigger Flooding predicted by a Gumstix.
Edge x
Edge x
Node D
Node E
Edge x
Node F

• Bluetooth continues to be used.
• Fewest Hop tree overlay applied to increase the
  robustness of the tree.

• Bluetooth used by default due to lower power
  requirements.
• Shortest-Path tree overlay used due to its power
  conservation characteristics.

17
Adaptation Scenario 3

• Node Submersion
• Likely that nodes will become submerged during
  flooding
• Want nodes to remain connected for as long as
  possible
• Switch network types when submersion is predicted
• Bluetooth -gt Wifi or Wifi -gt GPRS
• High power consumption and increased range

18
Adaptation Scenario 3 cont.
Fewest Hops
Fewest Hops
Trigger Submersion predicted by a Gumstix.

• Switch from Bluetooth to Wifi
• Same overlay type used.
• However, the different characteristics of Wifi
  causes a new topology to be created.

19
Current Status

• Small test-bed of nodes currently deployed
• Three Gumstix Nodes
• Depth Sensors
• Image-based Flow Sensors
• Additional nodes are being added, to initial
  deployment size of 13 nodes.
• Performance of network hardware, solar panels and
  other hardware is being tested in the field.

20
Future Work

• Development of a simulator to test deployment
  approaches with past and predicted flood events.
• Bringing in more highly embedded hardware running
  the RUNES GridKit implementation.
• Integration with Lancasters main NW Grid
  Deployment.

21
Summary

• Proposes a more automatic and sophisticated
  mechanism for collecting and processing flood
  data
• Convergence of Grid and Wireless Sensor Network
  functionality
• Uses this sophisticated mechanisms for performing
  adaptations
• Can customise the configuration and behaviour of
  sensors to the current environmental conditions