Multi-Objective Routing Optimisation for Battery

Transcription

Multi-Objective Routing Optimisation for
Battery-Powered Wireless Sensor Mesh Networks
Alma Rahat
Richard Everson
Jonathan Fieldsend
Computer Science
University of Exeter
United Kingdom
Genetic and Evolutionary Computation Conference, July 2014
Rahat, Everson & Fieldsend
MORO for Battery Powered WSMN
GECCO, July 2014
1 / 13
Wireless Sensors
Autonomous devices
Environmental or process
monitoring
Industrial
Heritage
Pharmaceuticals
Health-care
Battery powered
Monitor locations that are
difficult to access
Typically left unattended for
long periods of time
GECCO, July 2014
2 / 13
Point-to-Point Networks
Sensors and Gateway
GECCO, July 2014
3 / 13
Sensors and Gateway
Direct connections between
Sensor and Gateway
GECCO, July 2014
3 / 13
Sensors and Gateway
Sensor and Gateway
Challenges
Limited Range
GECCO, July 2014
3 / 13
Sensors and Gateway
Sensor and Gateway
Challenges
Limited Range
Vulnerable to dynamic radio
environment
GECCO, July 2014
3 / 13
Mesh Networks
Sensor nodes relay their
adjacent nodes’ data to the
gateway
GECCO, July 2014
4 / 13
Mesh Networks
gateway
Range extension
GECCO, July 2014
4 / 13
Mesh Networks
gateway
Range extension
Alternative routes - resilience
to changes in radio
environment
GECCO, July 2014
4 / 13
Mesh Networks
gateway
Range extension
to changes in radio
environment
GECCO, July 2014
4 / 13
Mesh Networks
gateway
Range extension
to changes in radio
environment
Maximise
Average battery lifetime
Minimum time before one node
expires
GECCO, July 2014
4 / 13
Routing Scheme
Network connectivity map
GECCO, July 2014
5 / 13
Routing Scheme
A route for node v3 :
S3 = hv3 , v1 , vG i
GECCO, July 2014
5 / 13
Routing Scheme
A route for node v3 :
S3 = hv3 , v1 , vG i
A routing scheme for the
network:
R = {S1 , S2 , S3 , S4 , S5 }
GECCO, July 2014
5 / 13
Node Costs
Node’s cost due to a routing
scheme R:
C1 = T1,G
+ (R1,2 + T1,G )
+ (R1,3 + T1,G )
For a transmission from vi to vj :
Ti,j Transmission cost at
node vi
Rj,i Reception cost at
node vj
GECCO, July 2014
6 / 13
Node Costs
scheme R:
C1 = T1,G
+ (R1,2 + T1,G )
+ (R1,3 + T1,G )
T1,G
node vi
node vj
GECCO, July 2014
6 / 13
Node Costs
scheme R:
C1 = T1,G
+ (R1,2 + T1,G )
+ (R1,3 + T1,G )
T1,G
R1,2
node vi
node vj
GECCO, July 2014
6 / 13
Node Costs
scheme R:
C1 = T1,G
R1,3
+ (R1,2 + T1,G )
+ (R1,3 + T1,G )
T1,G
node vi
node vj
GECCO, July 2014
6 / 13
Objectives
Lifetime for node vi :
Li (R) =
Qi
Ei + Ci
Qi battery charge
Ei quiescent current
Ci radio communication current
GECCO, July 2014
7 / 13
Objectives
Lifetime for node vi :
Li (R) =
Qi
Ei + Ci
Qi battery charge
Ei quiescent current
Ci radio communication current
Maximise
n
1X
Li (R)
n i=1
Average lifetime:
f1 (R) =
Minimum lifetime:
f2 (R) = min Li (R)
i∈[1,n]
GECCO, July 2014
7 / 13
Search Space Size
How big is the search space?
GECCO, July 2014
8 / 13
Search Space Size
Number of possible loopless
paths for node v3 : 1
GECCO, July 2014
8 / 13
Search Space Size
GECCO, July 2014
8 / 13
Search Space Size
GECCO, July 2014
8 / 13
Search Space Size
GECCO, July 2014
8 / 13
Search Space Size
GECCO, July 2014
8 / 13
Search Space Size
GECCO, July 2014
8 / 13
Search Space Size
GECCO, July 2014
8 / 13
Search Space Size
Number of possible routing
schemes:
n
Y
ai
i=1
ai : Number of available routes
from vi to vG
GECCO, July 2014
8 / 13
Search Space Size
schemes:
n
Y
ai
i=1
from vi to vG
4032 solutions
GECCO, July 2014
8 / 13
Search Space Size
schemes:
n
Y
ai
i=1
from vi to vG
243 solutions
GECCO, July 2014
8 / 13
Search Space Size
schemes:
n
Y
ai
Limit the number of paths
available to each node by using
k-shortest paths algorithm
[Yen, 1972; Eppstein, 1999]
Maximum search space size: k n
i=1
from vi to vG
Quicker approximation of
Pareto Front
243 solutions
GECCO, July 2014
8 / 13
Multi-Objective Evolutionary Algorithm
1:
2:
3:
4:
5:
6:
7:
8:
A ← InitialiseArchive()
. Initialise elite archive randomly
for i ← 1 : T do
R1 , R2 ← Select(A)
. Select two parent solutions
R0 ← UniformCrossOver (R1 , R2 )
R00 ← Mutate(R0 )
A ← NonDominated(A ∪ R00 )
. Update archive
end for
return A
. Approximation of the Pareto set
Crossover Select paths for each node from parents
Mutation Replace paths randomly from k-shortest paths for some
nodes
GECCO, July 2014
9 / 13
Real Network: The Victoria & Albert Museum
GECCO, July 2014
10 / 13
GECCO, July 2014
10 / 13
Minimum
Lifetime (years)
Minimum Lifetime (years)
1.2
30 nodes + gateway
k = 10; search space
is limited to 1030
solutions.
1.1
1.0
Initial population
size: 100
0.9
0.8
Mutation and
crossover rate: 0.1
0.7
0.6
1.86
1.88
1.90
1.92
1.94
1.96
Average Lifetime (years)
1.98
2.00
2.02
Number of iterations:
150, 000
Run time: 2 minutes
GECCO, July 2014
10 / 13
Minimum
Lifetime (years)
1.2
30 nodes + gateway
is limited to 1030
solutions.
1.1
1.0
Initial population
size: 100
0.9
0.8
Mutation and
crossover rate: 0.1
0.7
0.6
1.86
1.88
1.90
1.92
1.94
1.96
1.98
2.00
2.02
150, 000
Run time: 2 minutes
GECCO, July 2014
10 / 13
Minimum
Lifetime (years)
1.2
30 nodes + gateway
is limited to 1030
solutions.
1.1
1.0
Initial population
size: 100
0.9
0.8
Mutation and
crossover rate: 0.1
0.7
0.6
1.86
1.88
1.90
1.92
1.94
1.96
1.98
2.00
2.02
150, 000
Run time: 2 minutes
GECCO, July 2014
10 / 13
Average lifetime: 2 years
Minimum lifetime: 0.75 years (node v19 )
20
2.2
7
26
18
4
15
24
14
23
21 19
1
25
1.1
Min. Lifetime
16 30 17
6
2
0
11
0.9
0.7
1.92
Gateway
27
0.8
1.93
1.94
1.95
1.96
1.97
1.98
1.99
2.00
8
13
1.0
3
9
16
30 nodes + gateway
10
20
29 5
12
18
1.8
14
is limited1.6to 103012
10
solutions.
1.4
Initial population8
size: 1001.2
6
Edge Utilisation
28
2.0
Lifetime Remaining (years)
22
Mutation and
4
1.0
crossover rate: 0.1
2
Number 0.8
of iterations:
150, 000
2.01
Avg. Lifetime
Run time: 2 minutes
GECCO, July 2014
10 / 13
Average lifetime: 1.93 years
20
2.2
7
26
18
4
15
24
14
23
21 19
1
25
1.1
Min. Lifetime
16 30 17
6
2
0
11
0.9
0.7
1.92
Gateway
27
0.8
1.93
1.94
1.95
1.96
1.97
1.98
1.99
2.00
8
13
1.0
3
9
16
30 nodes + gateway
10
20
29 5
12
18
1.8
14
10
solutions.
1.4
Initial population8
size: 1001.2
6
Edge Utilisation
28
2.0
22
Mutation and
4
1.0
crossover rate: 0.1
2
Number 0.8
of iterations:
150, 000
2.01
Avg. Lifetime
Run time: 2 minutes
GECCO, July 2014
10 / 13
Average lifetime: 1.97 years
20
2.2
7
26
18
4
15
24
14
23
21 19
1
25
1.1
Min. Lifetime
16 30 17
6
2
0
11
0.9
0.7
1.92
Gateway
27
0.8
1.93
1.94
1.95
1.96
1.97
1.98
1.99
2.00
8
13
1.0
3
9
16
30 nodes + gateway
10
20
29 5
12
18
1.8
14
10
solutions.
1.4
Initial population8
size: 1001.2
6
Edge Utilisation
28
2.0
22
Mutation and
4
1.0
crossover rate: 0.1
2
Number 0.8
of iterations:
150, 000
2.01
Avg. Lifetime
Run time: 2 minutes
GECCO, July 2014
10 / 13
Recovering from Link Failure
Select operating point from estimated Pareto front
Simulate radio activity for 6 months
Simulate link failure
Minimum
Lifetime (years)
Minimum
Lifetime
(years)
1.1
1.0
0.9
0.8
0.7
1.92
1.93
1.94
1.95
1.96
1.97
1.98
AverageLifetime
Lifetime (years)(years)
Average
1.99
2.00
GECCO, July 2014
11 / 13
Select operating point from estimated Pareto front 3
Simulate radio activity for 6 months
2.2
7
22
28
1.0
15
4
23
0.9
14
25
2.0
26
18
10
20
21 19 29 5
12
8
2
16 30 17
24
1
6
13
0
27
3
1.6
1.4
1.2
1.0
11
0.8
1.8
Minimum
Lifetime (years)
Minimum
Lifetime
(years)
1.1
0.8
9
0.6
0.7
1.92
1.93
1.94
1.95
1.96
1.97
1.98
AverageLifetime
Average
1.99
2.00
GECCO, July 2014
11 / 13
Simulate radio activity for 6 months 3
1.0
7
22
28
0.8
15
4
23
25
0.6
10
20
21 19 29 5
12
8
2
16 30 17
24
1
6
13
0
27
0.0 months
1.5
1.6
1.7
1.8
1.9
AverageLifetime
Average
3
1.8
1.6
1.4
1.2
1.0
11
0.4
0.2
14
2.0
26
18
Minimum
Lifetime (years)
Minimum
Lifetime
(years)
2.2
0.8
9
0.6
2.0
GECCO, July 2014
11 / 13
Simulate radio activity for 6 months 3
Simulate link failure 3
1.0
7
22
28
0.8
15
4
23
25
0.6
10
20
21 19 29 5
12
8
2
16 30 17
24
1
6
13
0
27
6.0 months
1.5
1.6
1.7
1.8
1.9
AverageLifetime
Average
3
1.8
1.6
1.4
1.2
1.0
11
0.4
0.2
14
2.0
26
18
Minimum
Lifetime (years)
Minimum
Lifetime
(years)
2.2
0.8
9
0.6
2.0
GECCO, July 2014
11 / 13
Reoptimise with aged front
0.8
Minimum
(years)
MinimumLifetime
Lifetime (years)
0.7
Aged front
0.6
0.5
0.4
0.3
0.2
1.38
1.40
1.42
1.44
1.46
1.48
AverageLifetime
Lifetime (years)
Average
(years)
1.50
1.52
GECCO, July 2014
11 / 13
0.8
Minimum
(years)
MinimumLifetime
Lifetime (years)
0.7
Aged front
0.6
0.5
0.4
0.3
0.2
1.38
1.40
1.42
1.44
1.46
1.48
AverageLifetime
Lifetime (years)
Average
(years)
1.50
1.52
GECCO, July 2014
11 / 13
0.8
Minimum
(years)
MinimumLifetime
Lifetime (years)
0.7
Aged front
0.6
0.5
0.4
0.3
0.2
1.38
1.40
1.42
1.44
1.46
1.48
AverageLifetime
Lifetime (years)
Average
(years)
1.50
1.52
GECCO, July 2014
11 / 13
0.8
Minimum
(years)
MinimumLifetime
Lifetime (years)
0.7
Reoptimised front
Aged front
0.6
0.5
0.4
0.3
0.2
1.38
1.40
1.42
1.44
1.46
1.48
AverageLifetime
Lifetime (years)
Average
(years)
1.50
1.52
GECCO, July 2014
11 / 13
Extending Minimum Lifetime
GECCO, July 2014
12 / 13
Charge
Node 1
R1
Node 5
Time
GECCO, July 2014
12 / 13
Charge
Node 1
R2
Node 5
Time
GECCO, July 2014
12 / 13
R1
Charge
Node 1
R1
Node 5
Time
GECCO, July 2014
12 / 13
R1 + R2
Charge
Node 1
R2
Node 5
Time
GECCO, July 2014
12 / 13
1.3
1.2
1.1
hR1 , R2 , R3 i
1.0
0.9
0.8
0.7
0.6
1.84
1.86
1.88
1.90
1.92
1.94
1.96
1.98
2.00
2.02
GECCO, July 2014
12 / 13
1.3
1.2
1.1
1.0
0.9
0.8
0.7
0.6
1.84
1.86
1.88
1.90
1.92
1.94
1.96
1.98
2.00
2.02
GECCO, July 2014
12 / 13
1.3
1.2
1.1
1.0
0.9
0.8
0.7
0.6
1.84
1.86
1.88
1.90
1.92
1.94
1.96
1.98
2.00
2.02
GECCO, July 2014
12 / 13
1.3
e
1.2
2.25
14
23
24
1
25
6
26
21 19
1.0
2
20
29
12
1.80
27
1.65
13
3
6
0
25
2
15
14
23
10
16 30 17
24
21 19
1
6
27
20
5
29
12
1.80
8
1.65
13
3
9
0
2.25
1.50
2.10
1.35
1.95
1.20
26
18
25
0.7
1.35
7
22
28
4
21 19
1
1.95
10
16 30 17
24
1.50
11
9
0.8
14
23
8
2.10
26
18
15
4
5
0.9
11
28
1.95
10
16 30 17
2.25
7
22
15
4
2.10
2
1.20
18
7
1.1
28
22
20
29
12
5
1.80
8
1.65
13
0
1.50
11
27
0.6
1.84
3
1.86
9
1.35
1.88
1.90
1.20
1.92
1.94
1.96
1.98
2.00
2.02
GECCO, July 2014
12 / 13
1.3
Multiple Routing Scheme
1.2
1.1
1.0
0.9
0.8
Single Routing Scheme
0.7
0.6
1.84
1.86
1.88
1.90
1.92
1.94
1.96
1.98
2.00
2.02
GECCO, July 2014
12 / 13
Summary
Multi-objective optimisation of
routing schemes to extend
battery powered mesh network
lifetime
Novel k-shortest path search
space pruning enables rapid
optimisation
Dynamic reoptimisation allows
recovery from node or link
failure
Current Work
Novel temporal load balancing
to improve performance
Find optimum time span for
component routing schemes
Patent applied for with the
IMC Group Ltd.
Protect a group of nodes
GECCO, July 2014
13 / 13

Multi-Objective Routing Optimisation for Battery

Transcription

Similar documents