Proactive monitoring in natural environments Ian Marshall, Computing Laboratory, University of Kent Technical Director of the Envisense Research centre
Current research methods Single expensive package In situ process studies Low spatial resolution Short lifetime Small areas
Wireless Sensor networks Ad-hoc wireless communication Physical measurement No access to mains Large area (sq kms) Long life (months) Many measurement points
WSN management Low probability of manual intervention Highly dynamic, unpredictable environment Very unreliable nodes and comms Need to automate response to events model free adaptive control
Peak district Experiments
Floodnet
SECOAS Scroby sands wind farm and its impact on sedimentation processes
CEFAS Survey April 2002
Mechanical General Arrangement Buoy (yellow) Radio equipment Data cable Warp Chain Warp Plough anchor
Real trial Oct-Nov 2004
Initial Deployment Areas 1 NM 6 Sensors 150m apart Shore station
Seabed Package Measure Oceanographic variables (15 minute cycle) Temperature (1 sample/min) Pressure (1 sample/s for 5 mins) Turbidity (10 samples/min) Tilt (aka current) - (1 sample/s for 5 mins) Conductivity (1 sample/min) Adapt sampling rates Adaptively log data Transmit selected data to radio buoy
Adaptive sampling Measure, delete, combine, forward, sleep Use local variability, neighbour variability and internal state Self configure using distributed evolutionary algorithm (bacteria) Can adjust priorities and frequency of actions Can form groups (quorum sensing) Reward set by user using a diffusion (gossip) protocol – changes drive auto- reconfiguration of genome
QoS on a Sensor Network
Processing
Summary Autonomous adaptive control is needed in environmental sensor networks Network protocols must support and respond to application semantics (be app aware) In simulation adaptation was almost as good as optimal sliding window In practice it dealt well with change from calm to stormy More research will be needed