changsenxia/-wsn-

Simple Wireless-Sensor-Network Optimization using Genetic-Algorithm.

WSN-Optimization-using-GA

Sensor placement in a network using Genetic Algorithm properties.

What's Genetic Algorithm?

GA is an algorithm inspierd from Darwin's theory of natural evolution and the mechanism of natural selection. Consider a society (population) contains arbitrary number of individuals; Two individuals (parents) bred and had two children they would have the genes and features of their parents with the probability of mutation (have extra feature or abandon inherited one). Children would grow and became a part of the society (population) and repeat the process for fittest individuals (highest scores) and have higher ability to servive and pass their genes or features to thier progeny.

How GA simulates natural selection mechanism?

• At first, We start with initial random population.
• Fittest ones (highest scores) have higher probability to live and breed.
• Each couple from fittest ones (parents) breed and produce children with mixed features from both of them. (see CrossOver.jpg)
• Weak individuals extinects and replaced with better generation.
• repeat the process untill the goal reached.

Pseudo Code:

population <- Random Population
Repeat N times or untill goal is reached:
    scores <- calculate_score_of(individual) For all individuals in population
    selected <- list of fittest individuals in population
    for parent1, parent2 in seleted:
        child1, child2 <- breedAndCrossOver(parent1, parent2)
        with mutationRate probability:
          changeFeatureFor(child1 or child2)
        newPopulation.add(child1)
        newPopulation.add(child2)
    population <- newPopulation

setup

Make sure that numpy and tkinter are installed. Use pip install numpy and pip install tkinter; if they're not installed they will be downloaded and installed. Then, run python Runner.py 100 100 20 20 you will be able to see something like this:

Project specfication and description:

Applying Genetic Algorithm concepts to find the best placement for sensors in Sensor netowrk. Each sensor had position (X, Y) and radius descriping its domain. We represent sensor as chromosome or string of features to manipulate. After all the process is done we can visulaize generations to see how it did.

Code Walkthrough and Logic

There're three main Modules:

• GeneticAlgorithm.py: Which contains all the logic for genetic algorithm,
• CoverageArea.py: to calculate the objective function or fitness score ,
• DrawSensors.py: handles GUI and buttons to display sensors,
  And driver code:
• Runner.py: firing the GeneticAlgorithm Module and call DrawSensors with GA object.

GeneticAlgorithm:

There's main class GeneticAlgorithm has bunch of attributes: iterationsNo or generation number, populationNo population size, bitsNo number of bits per individual in population, cross_over_rate can be considered as the probability to have a crossover, mutation_rate probability for mutation to occure, sensors_radius the radius of the sensor caoverage, population sensors locations, best_gen fittest generation index, best_score highest covered area so far, generations list of populations (society within an era) and finaly, coverage list of total coverage areas per generation. And some methods such as init_population which initiate population, objective to calculate the fitness score, select to select fit individual, cross_over to reproduce two children from two selected parents, mutation to mutate random feature for child and main method start which fires logic by calling all of them.

As I discussed about Genetic Algorithm we apply steps as follow:

• Initiate random population.
• Record coverage area for sensors (fitness score).
• Select two fit parents to breed and reproduce two children have traits of them with probability to mutate.
• Do all of above for all parents untill the compeletion for generation then start with last population.

CoverageArea:

Simple module to calculate the total coverage area for sensors and corresponding area for each one to calculate the fitness score by traversing all points (pixels) in the plane and count covered one.

DrawSensors:

GUI tkinter window to display sensors as generations given GeneticAlgorithm object GA. See this if unfamiliar with tkinter.

Notes:

• This project may not be the most suitable or direct application of Genetic Algorithm but nice start to understand.
• Perfomance is pretty bad so, enhancements are coming soon.

Refrences:

Machine Learning Mastery article from the fascinating Jason Brownlee.