Route Selection of Mobile Sensors for Air Quality Monitoring

Transcription

1 Route Selection of Mobile Sensors for Air Quality Monitoring Olga Saukh, David Hasenfratz, Abouzar Noori, Tamara Ulrich, and Lothar Thiele Computer Engineering and Networks Laboratory ETH Zurich, Switzerland PerSeNS 2012 March 19, Lugano, Switzerland

2 Today Sensing the Air We Breathe Air pollution is a major concern in urban areas 2

3 Static Monitoring Networks Governmental monitoring networks Managed by official authorities Analytical instruments accurately measure wide ranges of air pollutants Cost, size and laborious maintenance severely limits the number of stations OstLuft stations, Balsberg & Winterthur NABEL station, Duebendorf 3

4 Air Pollution Maps NABEL network 16 stations all over Switzerland, 1 station in Zurich 8 different zones, air pollution maps are imprecise OstLuft cantonal network 34 stations in east cantons (only main pollutants), 4 stations in Zurich NABEL measurement station NABEL network Ozone map GOAL: Improve the resolution of current air pollution maps in cities with lightweight OpenSense sensing stations 4

5 Mobile OpenSense Network 5

6 OpenSense Nodes 1 station (March: 5 stations, end 2012: 10 stations): Sensors: O3, CO, particulate matter (PM), temperature, humidity, accelerometer GPS Communication: WLAN, Ethernet and GSM External power supply (tram) Powerful embedded Linux Gumstix, 600MHz CPU 128 MB RAM/32 MB Flash Interfaces Power-switchable USB Analog/digital 6

7 Mobile OpenSense Nodes OpenSense nodes are installed on top of several public transport vehicles Zurich public transport (VBZ) operates 14 tram and 54 bus lines 15 vehicles on average are operating on each of the 68 lines during the day PROBLEM: Which routes to pick for installation of OpenSense nodes to cover the city well? 7

8 Coverage of a City Assumptions In most cases mobile vehicles follow their designated tracks and timetables with no delays or arrivals ahead of time All exceptions are infrequent Problem Area of interest Ω Timetable network N = (H, S, C) Set of mobile vehicles H, set of nodes S, and set of elementary connections C Set of nodes K An elementary connection is a 5-tuple c = (h, s1, s2, t1, t2) Density requirement function ρ: Ω (0, 1] represents the measurement density demand in the area Ω Validity function of a measurement at a point in its vicinity w : Ω x Ω [0, 1] Coverage ( H ) ( x) max w( z, x)dx K z h, h H K 8

9 Checkpoints Two vehicles make a checkpoint if the distance between them is below a certain threshold. Checkpoints are useful for: Relating measurements in space and time Comparing sensor readings and sensor calibration Recognizing faulty sensors Types of checkpoints: X-Checkpoint - between two OpenSense nodes R-Checkpoint - between an OpenSense node and a reference station 9

10 Checkpointing Constraints Given checkpoints between all pairs of sensors, it is possible to construct a checkpoint graph If a checkpoint graph is connected, the set of selected vehicles fulfills checkpointing constraints. X-Checkpointing 10

11 Checkpointing Constraints Given checkpoints between all pairs of sensors, it is possible to construct a checkpoint graph If a checkpoint graph is connected, the set of selected vehicles fulfills checkpointing constraints. R-Checkpointing 11

12 Route Selection Problem Route selection problem involves high computational complexity even for a small number of OpenSense nodes The brute force approach would require to go through K combinations to compute the optimum timetable subnetwork of size K H 12

13 Chromosome Evolutionary Algorithm K Tram 1, 10:48 Tram 2, 10:55 Tram 5, 10:03 Tram 1, 11:13 Generation 60 K K K K Fitness function is inverse to coverage 13

14 Evolutionary Algorithm Crossover Each pair is selected with 70% probability 50% mixing rate + Mutation Replacing random vehicle with a new one Repeated to satisfy constraints, max. 5 times Selection Decides which chromosomes of the parents and offsprings are going to survive Restricted tournament: A variated chromosome can only substitute its more similar parent Better fitness value wins 14

15 Scenarios Coverage-only optimization Reliable and trusted system and sensors No need to cross-check the measurements Coverage with X-checkpointing constraints Unreliable or untrusted systems and sensors Small cities might not have any reference station Coverage with R-checkpointing constraints Unreliable or untrusted systems and sensors Big Cities with a few reference stations 15

16 One-Dimensional City Space Time 16

17 One-Dimensional City Coverage-only Coverage with X-checkpointing Coverage with R-checkpointing 17

18 Evaluation: City of Zurich Tram Network of Zurich City: 2 reference stations in the city center 13 tam lines 10 to 20 trams serving one line 260 trams in total Evaluation Setup: Checkpoint - within 200m 3 Scenarios Coverage-only Coverage with X-Checkpointing Coverage with R-Checkpointing Random Search Simulated Annealing Evolutionary Algorithm 20 times 18

19 Simulated Annealing & Random Search Random Search Generate a large number of solutions and pick the best one Simulated Annealing Start with initial configuration Repeatedly search neighborhood and select a neighbor as candidate Evaluate fitness function and accept candidate if better; if not, select another neighbor Do sometimes accept candidates with higher cost to escape from local optimum Adapt the parameters of the evaluation function during execution Based upon the analogy with the simulation of the annealing of solids Stop if quality is sufficiently high, if no improvement can be found or after some fixed time 19

20 better Zurich City, 2 trams Evolutionary Algorithm Simulated Annealing Random Search 20

21 better Zurich City, 10 trams Evolutionary Algorithm Simulated Annealing Random Search 21

22 Conclusions Air pollution is of major concern in modern cities High resolution maps of distribution of air pollutants in cities with the OpenSense network Route selection to cover a city well Checkpointing constraints to calibrate the sensors and to recognize faulty sensors Evolutionary algorithm for route selection Evaluation based on the tram network of Zurich Ozone concentration in Zurich (source: data.opensense.ethz.ch) 22

23 Interested in Air Pollution? 23