Docstoc

ENERGY-AWARE GENETIC ALGORITHM FOR MANAGING WIRELESS SENSOR NETWORKS

Document Sample
ENERGY-AWARE GENETIC ALGORITHM FOR MANAGING WIRELESS SENSOR NETWORKS Powered By Docstoc
					 Energy-Aware Genetic
Algorithm for Managing
Wireless Sensor Networks

            Abhishek Karpate
Overview
§ Wireless Sensor Networks are quickly emerging
  as a technology for tracking and monitoring in
  many domains.
  §   Military applications
  §   Environmental applications
  §   Health applications
  §   Agricultural applications
  §   Smoke detection
§ They consist of spatially distributed sensors
  which cooperate among themselves for any task.
§ The sensors are small integrated circuits which
  are embedded in wireless devices.
Challenges
• Reliability and Robustness
   Sensor networks are not meant for frequent maintenance.
   They should be operated by     sensors that are reliable and
   should be deployed in large numbers.
• Energy conservation
   Sensor networks have limited computational and
   communication capabilities. Algorithms need to be
   collaborative with energy aware communication.
• Real time data acquisition and processing
   There is a critical need for efficient data communication and
   data processing. Some techniques that are used are event
   ordering and synchronization.
• Data Management
   An embedded real time database is needed which stores the
   data of interest and provides results to different queries.
• Data privacy and Security
   The data collected is sensitive. It should be made sure that
   the data is properly transmitted and collected with no loss.
Previous Work
 • Many Researchers approached the management
   of wireless sensors problem from one specific
   angle such as network life span and coverage.
 • Very little has been done to optimize multiple
   objectives concurrently which is critical in this
   case, in particular the trade off between
   performance and consumed energy.
 • Almost all current approaches provide static
   solutions that lack the flexibility to change
   priority according to the application domain.
 • Few research studies looks at the impact of the
   density of the sensor nodes.
Approach




                                           Area monitored by
                                           single sensor
                                           Area monitored by
                                           multiple sensors

Set of Sensor Nodes   Set of Sub-regions
The Proposed Model
WSN and the set of areas it needs to cover can be
represented by a bipartite graph G as follows:
                 G = ( N ∪ A, E); where,
    N  Set of sensor nodes
    A  Set of area monitored
    E  Edge which exists between a node in N to a node in
    A if and only if that particular node monitors that area.
Key Features of the Model
  § The proposed model provides a flexible
    mechanism to incorporate various parameters
    of the problem, such as:
    § Heterogeneity of the sensor network
    § Optimization criteria
  § An any given instance, a solution is given by
    identifying a subset of nodes representing
    sensors (active sensors) that dominates the
    entire set of nodes representing the areas need
    to be covered.
  § Such a subset needs to be identified in a way
    that attempts to optimize various parameters
    related to performance and energy.
Evolutionary Algorithm
§ The evolution starts with n randomly generated
  strings.
§ The length of the string represents the number of
  sensors.
§ The state of each individual sensor is represented by a
  1bit binary number called gene.
§ The gene defines the status of the nodes as follows:

             si =
Mutation
§ Randomly generated strings are subjected to
  mutation.
§ Mutation is the random alteration of a gene of a
  chromosome.
§ Helps in reintroducing the random cells that have
  been lost.
  Original String:
        1 0 1 0 1 1 0 0 0 1 0 0 0 1 0 0 1 0 1
  Mutated String:
       1 0 1 0 1 1 1 0 0 1 0 1 0 1 0 0 1 0 0
Crossover
• Crossover is an operator that combines two strings to produce a
  new string.
• The purpose behind crossover is that the resulting string takes
  better characteristics from both the parents
        1 0 0 1 0 1 1 0 1 1 0 0 0 1 0 0 0 1 0 0 1 0 1

        1 1 0 0 0 1 1 0 0 0 1 1 0 1 0 1 1 1 0 0 0 0 0

• The length of each substring to be swapped is taken as input
  from the user.
• The number of strings generated depends on the length of the
  substring.




    Crossovers generated when the length of the substring is one-third the string length
Fitness Function
 As discussed there are the following sets and constants:
      S = {s1, s2, …, sn} the set of sensors
      A = {A1, A2, …, Am} the set of sub-regions

  We define fitness function, f ∀S, ∀A as follows:
                       f = αP + βQ + γR
 where,
 P  Percentage of coverage
 Q  Quality of coverage
 R  Number of inactive nodes
 and
 α, β, γ are tuning parameters to customize the function
 according to the priority of each application
Selection
 § The process of selecting better individuals to use as
   parents for the further generations.
 § The process of selection is stopped when for a few
   cycles no strings with better fitness are found.
      110101001   000101001
      100010001   100110011
      000101101   111101101
      001011100   111011100
                                     The healthiest
      111010001   111010001            strings are
      110101011   000101111
      101010111   100010111          considered for
                                   future generations
 § Elitism can occur in one of the following forms:
    • Pure Elitism: The strings that are considered are
       all obtained from the previous generation
    • Partial Elitism: Some of the strings that are
       considered are obtained from the previous
       generation, others are randomly generated.
Assessment
• A benchmark is needed to compare the
  performance of the algorithm.
• The performance of the algorithm is compared
  with the greedy Round-Robin algorithm.
• The algorithm is executed as follows:
  • The desired number of sensors that need to be active
    are taken as input.
  • Accordingly sensors are divided into groups.
  • For any particular time slice a particular group of
    sensors are turned on.
 Experiments
• The simulations are carried out on three types of
  networks – densely, medium and sparsely
  populated
• The simulations are carried out in order to achieve
  the following:
  • Have a set of options available for any desired situation.
  • As needed, the required numbers of sensors are turned
    on and the requirement is fulfilled.
  • Control the activation and deactivation of the sensors.
  • A network that swings its main priority from coverage
    to energy saved.
 Results
    Densely populated network
    Sensor networks having more than 50 sensors are classified as dense
    networks.


    Table 1: Simulation test results using genetic algorithm             Table 2: Simulation test results using round-robin algorithm

                 Input Parameters              Output Parameters                      Number of      %        Quality   Energy
                                                                                        Active    Coverage   Coverage   Saved
            %        Quality    Energy       %        Quality   Energy
                                                                                       Sensors
         Coverage   Coverage    Saved     Coverage   Coverage   Saved
                                                                            Case 1       40        91.35      49.07     33.33
Case 1     0.9         0.8          0.2    98.76      77.16     33.33
                                                                            Case 2       25        86.41      42.90     56.66
Case 2     0.8         0.5          0.5    96.29      73.14     56.66
                                                                            Case 3       18        81.48      47.53     71.66
Case 3     0.7         0.3          0.6    95.06      66.97     71.66
                                                                            Case 4       10        70.37      48.46     83.33
Case 4     0.4         0.2          0.8    88.88      72.53     83.33

Case 5     0.1         0.1          0.9    60.49      47.84     91.66       Case 5        5        49.38      39.82     91.66
    Intermediately populated network
    Sensor networks having less than 50 sensors and more than 25 sensors are
    classified as intermediately populated networks.

 Table 3: Simulation test results using genetic algorithm                 Table 4: Simulation test results using round-robin algorithm

                 Input Parameters              Output Parameters                       Number of      %        Quality   Energy
                                                                                         Active    Coverage   Coverage   Saved
            %        Quality    Energy       %        Quality   Energy
                                                                                        Sensors
         Coverage   Coverage    Saved     Coverage   Coverage   Saved
                                                                             Case 1       23        81.48      46.29      42.5

Case 1     0.9         0.8          0.2    93.83      79.94        42.5      Case 2       16        76.54      45.68      60.0
Case 2     0.8         0.5          0.5    91.36      70.68        60.0
                                                                             Case 3       13        71.61      51.23      67.5
Case 3     0.7         0.3          0.6    88.88      61.42        67.5
                                                                             Case 4        8        55.55      33.03      82.5
Case 4     0.4         0.2          0.8    81.48      64.19        82.5
Case 5     0.1         0.1          0.9    27.16      27.16        95.0      Case 5        2        17.28      17.28      95.0
Sparsely populated network
Sensor networks having less than 25 sensors are classified as sparsely
populated networks.

Table 5: Simulation test results using genetic algorithm                  Table 6: Simulation test results using round-robin algorithm

                 Input Parameters              Output Parameters                        Number of      %        Quality   Energy
                                                                                          Active    Coverage   Coverage   Saved
            %        Quality    Energy       %        Quality   Energy
                                                                                         Sensors
         Coverage   Coverage    Saved     Coverage   Coverage   Saved
                                                                              Case 1       18        85.18      52.16      28.0
Case 1     0.9         0.8          0.2    90.12      71.30        28.0
                                                                              Case 2       11        67.90      29.32      56.0
Case 2     0.8         0.5          0.5    81.48      52.16        56.0
                                                                              Case 3        8        54.32      32.72      68.0
Case 3     0.7         0.3          0.6    75.31      60.18        68.0
                                                                              Case 4        5        46.91      38.89      80.0
Case 4     0.4         0.2          0.8    58.02      47.84        80.0

Case 5     0.2         0.1          0.9    32.01      32.01        92.0       Case 5        2        14.81      12.96      92.0
Analysis of Results
Ø Genetic algorithms overall provide better coverage for the same
  amount of energy used as compared to the round-robin algorithm.

Ø A better quality of coverage is attained by spending equal amount
  of energy in genetic algorithms which provide a much needed
  redundancy for critical applications.

Ø The more densely populated a network is the better capable it is in
  producing a wider range of viable results.

Ø As the network goes more sparse, lower amounts of energy would
  be saved in order to attain the some level of coverage or quality.

Ø The more number of strings considered in the partial elitism the
  better are the chances of getting more viable results.

Ø More efficient results are attained if the number of elements in the
  crossover operation are kept low because then the crossover is
  capable of producing more offspring.
Conclusions
Ø The project proposes a scheme by which the
  activation/deactivation process in a wireless sensor
  network can be adjusted to optimize multiple objectives.

Ø According to the specific need at a given instance, the
  behavior of the wireless sensor network can be changed
  depicting networks of different priorities.

Ø The proposed evolutionary algorithm presents different
  solutions to the optimization problem and outperforms
  standard greedy techniques in finding the most fit solution.

Ø The graph theoretic model along with the algorithm can
  be further expanded by creating strings that represent
  various states of the sensors such as transmission, listening,
  active, sleep.
Thank You

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:13
posted:8/5/2013
language:English
pages:20
Lingjuan Ma Lingjuan Ma MS
About work for China Compulsory Certification. Some of the documents come from Internet, if you hold the copyright please contact me by huangcaijin@sohu.com