Cooperative Information Augmentation in a Geosensor Network

Transcription

1 Cooperative Information Augmentation in a Geosensor Network Malte Jan chulze, Claus Brenner & Monika ester Institute of Cartography and Geoinformatics Leibniz University Hannover

2 Outline Introduction Properties of Geosensor Networks Problem statement imulation environment Basic concept Car movement Rainfall simulation Observation and mapping of rainfall Experiments Alteration of car density Conclusion and future work 2

3 Introduction

4 Properties of Geosensor Networks Classical configuration: ensors are distributed manually Centralized organization of sensors and computation of measurement data High data volume and computational costs in large networks Installation of sensors and data communication requires complex and expensive planning tatic system Base 4

5 Properties of Geosensor Networks Geosensor Network: ensor node possesses processor, memory, battery and wireless communication interface Local processing with parallel algorithms Coordination with neighbors to reach global knowledge Network access via sinks Decentralized and self-organizing system calable High redundancy Fault tolerance Dynamic system ink 5

6 Problem statement Rainfall is important source for hydrological planning and water resource management Modeling of high dynamic processes like floods and erosion rely on high resolution rainfall information Density of recording weather stations is too small (e.g. in Germany 1 station per 1800 km²) Idea: Densify the number of stations with massively available, unconventional and cheap sensors 6

7 Problem statement unshine Wipers off Light rainfall Wipers min. speed Heavy rainfall Wipers max. speed Wiper frequency is an indicator for rainfall conditions Can be recorded by any type of car Functional dependency between wiper frequency and rainfall intensity needs to be determined for every car individually 7

8 imulation environment

9 Basic concept Describe quality of rain measurement using cars as rain gauges Car as moving sensor node Determine position (via GP) and wiper frequency Perform calculations based on locally collected data hare data with other cars using wireless communication Relationship between wiper frequency and rainfall is initially unknown Uncertainty of rainfall estimation very high Distribution of few weather stations with wireless communication Provide high quality rainfall measurements Cars improve own certainty of rainfall measurement with information from weather stations and from other cars with a higher certainty 9

10 Basic concept Communication strategy: tation b tation a Car C 1 10

11 Basic concept Communication strategy: tation b tation a Car C 1 Car C 2 11

12 Basic concept Communication strategy: σ I. Poor certainty due to unknown relationship II. III. Improvement of certainty while communicating with station/other car Certainty decreases slightly position station range 12

13 Car movement Each car follows a certain trajectory in road network tart and destination points picked randomly Calculation of path via A*- algorithm Record visited nodes with timestamp based on a given average speed imulation based on central start and end time using constant time steps of 10 s Linear interpolation of position between two nodes 13

14 Rainfall simulation Model of rainfall intensity um of mixed Gaussians with randomly distributed centers Normalization of whole field Rainclouds considered stationary mm 2 m s cloud( x, y) 1 = e 2πσ ( x x ) + ( y y ) σ 14

15 Rainfall observation Describe system state and quality Kalman filter implemented for every car x& +, 0 k + σ xk & σ wx& 0 xk = xx, k, + Σ = + 2 Σww = 2 x&& 0 σ k xk &&, 0 σ wx&& k k k+ 1 1 Δt k xk+ 1 = x& k +Δtk x&& k Φk = 0 1 Rainfall intensity [mm/m²s] Change of rainfall intensity [mm/m²s²] + Covariance matrix of system state Σ > Increases with time as system noise accumulates Prediction with transition matrix Φ x& Quality of rainfall measurement described by x&& xx, k Σ ww σ + x&, k x, Σ + + k xx, k 15

16 Rainfall observation Update system state in three cases of communication I. No station or car in communication range Measure rainfall intensity with poor accuracy II. Weather station in communication range Receive rainfall intensity with high accuracy from station Ignore data from other cars in range III.Only other car(s) in communication range Receive system state with standard deviation from other car(s) Perform update of own system state Use received/measured data as observation(s) 16

17 Rainfall mapping Convert road network to raster data Cell including part of road network is possible candidate to receive rainfall information from car Each cell of road network carries Kalman filer x& Σ = σ, Σ = σ Φ k xxk, xk &, ww wx& k+ 1 k = 1 Cell is updated when visited by a car using system state and quality ystem noise models decay in quality x&, σ + + k x&, k 17

18 Experiments

19 Experiments Improvement of certainty due to communication imulation with 50 cars tandard deviation of rainfall measurement 6 % 25 % of cells mapped Cell visiting rate at

20 Experiments tandard deviation of reached cells Number of visits weather station weather station mm 2 m s # Very good mapping quality around stations Highly correlated with number of visits

21 Experiments imulation with 100 cars td. of rainfall measurement 7 % 35 % of cells mapped Cell visiting rate at 1.40 imulation with 50 cars td. of rainfall measurement 6 % 25 % of cells mapped Cell visiting rate at 0.69 weather station weather station mm 2 m s 21

22 Conclusion and future work Approach for densification of rainfall measurement via Geosensor Network Implementation of an agent based simulation environment Inaccurate estimation of rainfall from wiper frequencies by cars while moving Decentralized information processing using Kalman filters Evaluation of mapping quality based on standard deviation of grid cells ystem improvements Verification of ystem noise behaviour based on real data Implementation of moving and varying rain field Usage of real traffic data (density and speed) Derivation of functional relationship between wiper frequency and rainfall intensity 22

23 Future work Estimation of functional relationship Coarse estimation due to wiper frequency interval tandard deviation of 16 % Alternative: Cars with moisture sensors for wiper automatic 23

24 Thank you very much for your attention! Malte Jan chulze Institute of Cartography and Geoinformatics Leibniz University Hannover