Strength Pareto Evolutionary Algorithm

Strength Pareto Evolutionary Algorithm, SPEA, SPEA2.

Taxonomy

Strength Pareto Evolutionary Algorithm is a Multiple Objective Optimization (MOO) algorithm and an Evolutionary Algorithm from the field of Evolutionary Computation. It belongs to the field of Evolutionary Multiple Objective (EMO) algorithms. Refer to for more information and references on Multiple Objective Optimization. Strength Pareto Evolutionary Algorithm is an extension of the Genetic Algorithm for multiple objective optimization problems. It is related to sibling Evolutionary Algorithms such as Non-dominated Sorting Genetic Algorithm (NSGA), Vector-Evaluated Genetic Algorithm (VEGA), and Pareto Archived Evolution Strategy (PAES). There are two versions of SPEA, the original SPEA algorithm and the extension SPEA2. Additional extensions include SPEA+ and iSPEA.

Strategy

The objective of the algorithm is to locate and and maintain a front of non-dominated solutions, ideally a set of Pareto optimal solutions. This is achieved by using an evolutionary process (with surrogate procedures for genetic recombination and mutation) to explore the search space, and a selection process that uses a combination of the degree to which a candidate solution is dominated (strength) and an estimation of density of the Pareto front as an assigned fitness. An archive of the non-dominated set is maintained separate from the population of candidate solutions used in the evolutionary process, providing a form of elitism.

Procedure

Algorithm (below) provides a pseudocode listing of the Strength Pareto Evolutionary Algorithm 2 (SPEA2) for minimizing a cost function. The CalculateRawFitness function calculates the raw fitness as the sum of the strength values of the solutions that dominate a given candidate, where strength is the number of solutions that a give solution dominate. The CandidateDensity function estimates the density of an area of the Pareto front as where is the Euclidean distance of the objective values between a given solution the $k$th nearest neighbor of the solution, and is the square root of the size of the population and archive combined. The PopulateWithRemainingBest function iteratively fills the archive with the remaining candidate solutions in order of fitness. The RemoveMostSimilar function truncates the archive population removing those members with the smallest values as calculated against the archive. The SelectParents function selects parents from a population using a Genetic Algorithm selection method such as binary tournament selection. The CrossoverAndMutation function performs the crossover and mutation genetic operators from the Genetic Algorithm.

Heuristics

• SPEA was designed for and is suited to combinatorial and continuous function multiple objective optimization problem instances.
• A binary representation can be used for continuous function optimization problems in conjunction with classical genetic operators such as one-point crossover and point mutation.
• A value of 1 may be used for efficiency whilst still providing useful results.
• The size of the archive is commonly smaller than the size of the population.
• There is a lot of room for implementation optimization in density and Pareto dominance calculations.

Code Listing

Listing (below) provides an example of the Strength Pareto Evolutionary Algorithm 2 (SPEA2) implemented in the Ruby Programming Language. The demonstration problem is an instance of continuous multiple objective function optimization called SCH (problem one in [Deb2002]). The problem seeks the minimum of two functions: and , and . The optimal solutions for this function are . The algorithm is an implementation of SPEA2 based on the presentation by Zitzler, Laumanns, and Thiele [Zitzler2002]. The algorithm uses a binary string representation (16 bits per objective function parameter) that is decoded and rescaled to the function domain. The implementation uses a uniform crossover operator and point mutations with a fixed mutation rate of , where is the number of bits in a solution's binary string.

def objective1(vector)
  return vector.inject(0.0) {|sum, x| sum + (x**2.0)}
end

def objective2(vector)
  return vector.inject(0.0) {|sum, x| sum + ((x-2.0)**2.0)}
end

def decode(bitstring, search_space, bits_per_param)
  vector = []
  search_space.each_with_index do |bounds, i|
    off, sum = i*bits_per_param, 0.0
    param = bitstring[off...(off+bits_per_param)].reverse
    param.size.times do |j|
      sum += ((param[j].chr=='1') ? 1.0 : 0.0) * (2.0 ** j.to_f)
    end
    min, max = bounds
    vector << min + ((max-min)/((2.0**bits_per_param.to_f)-1.0)) * sum
  end
  return vector
end

def point_mutation(bitstring, rate=1.0/bitstring.size)
  child = ""
   bitstring.size.times do |i|
     bit = bitstring[i].chr
     child << ((rand()<rate) ? ((bit=='1') ? "0" : "1") : bit)
  end
  return child
end

def binary_tournament(pop)
  i, j = rand(pop.size), rand(pop.size)
  j = rand(pop.size) while j==i
  return (pop[i][:fitness] < pop[j][:fitness]) ? pop[i] : pop[j]
end

def crossover(parent1, parent2, rate)
  return ""+parent1 if rand()>=rate
  child = ""
  parent1.size.times do |i|
    child << ((rand()<0.5) ? parent1[i].chr : parent2[i].chr)
  end
  return child
end

def reproduce(selected, pop_size, p_cross)
  children = []
  selected.each_with_index do |p1, i|
    p2 = (i.modulo(2)==0) ? selected[i+1] : selected[i-1]
    p2 = selected[0] if i == selected.size-1
    child = {}
    child[:bitstring] = crossover(p1[:bitstring], p2[:bitstring], p_cross)
    child[:bitstring] = point_mutation(child[:bitstring])
    children << child
    break if children.size >= pop_size
  end
  return children
end

def random_bitstring(num_bits)
  return (0...num_bits).inject(""){|s,i| s<<((rand<0.5) ? "1" : "0")}
end

def calculate_objectives(pop, search_space, bits_per_param)
  pop.each do |p|
    p[:vector] = decode(p[:bitstring], search_space, bits_per_param)
    p[:objectives] = []
    p[:objectives] << objective1(p[:vector])
    p[:objectives] << objective2(p[:vector])
  end
end

def dominates?(p1, p2)
  p1[:objectives].each_index do |i|
    return false if p1[:objectives][i] > p2[:objectives][i]
  end
  return true
end

def weighted_sum(x)
  return x[:objectives].inject(0.0) {|sum, x| sum+x}
end

def euclidean_distance(c1, c2)
  sum = 0.0
  c1.each_index {|i| sum += (c1[i]-c2[i])**2.0}
  return Math.sqrt(sum)
end

def calculate_dominated(pop)
  pop.each do |p1|
    p1[:dom_set] = pop.select {|p2| p1!=p2 and dominates?(p1, p2) }
  end
end

def calculate_raw_fitness(p1, pop)
  return pop.inject(0.0) do |sum, p2|
    (dominates?(p2, p1)) ? sum + p2[:dom_set].size.to_f : sum
  end
end

def calculate_density(p1, pop)
  pop.each do |p2|
    p2[:dist] = euclidean_distance(p1[:objectives], p2[:objectives])
  end
  list = pop.sort{|x,y| x[:dist]<=>y[:dist]}
  k = Math.sqrt(pop.size).to_i
  return 1.0 / (list[k][:dist] + 2.0)
end

def calculate_fitness(pop, archive, search_space, bits_per_param)
  calculate_objectives(pop, search_space, bits_per_param)
  union = archive + pop
  calculate_dominated(union)
  union.each do |p|
    p[:raw_fitness] = calculate_raw_fitness(p, union)
    p[:density] = calculate_density(p, union)
    p[:fitness] = p[:raw_fitness] + p[:density]
  end
end

def environmental_selection(pop, archive, archive_size)
  union = archive + pop
  environment = union.select {|p| p[:fitness]<1.0}
  if environment.size < archive_size
    union.sort!{|x,y| x[:fitness]<=>y[:fitness]}
    union.each do |p|
      environment << p if p[:fitness] >= 1.0
      break if environment.size >= archive_size
    end
  elsif environment.size > archive_size
    begin
      k = Math.sqrt(environment.size).to_i
      environment.each do |p1|
        environment.each do |p2|
          p2[:dist] = euclidean_distance(p1[:objectives], p2[:objectives])
        end
        list = environment.sort{|x,y| x[:dist]<=>y[:dist]}
        p1[:density] = list[k][:dist]
      end
      environment.sort!{|x,y| x[:density]<=>y[:density]}
      environment.shift
    end until environment.size <= archive_size
  end
  return environment
end

def search(search_space, max_gens, pop_size, archive_size, p_cross, bits_per_param=16)
  pop = Array.new(pop_size) do |i|
    {:bitstring=>random_bitstring(search_space.size*bits_per_param)}
  end
  gen, archive = 0, []
  begin
    calculate_fitness(pop, archive, search_space, bits_per_param)
    archive = environmental_selection(pop, archive, archive_size)
    best = archive.sort{|x,y| weighted_sum(x)<=>weighted_sum(y)}.first
    puts ">gen=#{gen}, objs=#{best[:objectives].join(', ')}"
    break if gen >= max_gens
    selected = Array.new(pop_size){binary_tournament(archive)}
    pop = reproduce(selected, pop_size, p_cross)
    gen += 1
  end while true
  return archive
end

if __FILE__ == $0
  # problem configuration
  problem_size = 1
  search_space = Array.new(problem_size) {|i| [-10, 10]}
  # algorithm configuration
  max_gens = 50
  pop_size = 80
  archive_size = 40
  p_cross = 0.90
  # execute the algorithm
  pop = search(search_space, max_gens, pop_size, archive_size, p_cross)
  puts "done!"
end

References

Primary Sources

Zitzler and Thiele introduced the Strength Pareto Evolutionary Algorithm as a technical report on a multiple objective optimization algorithm with elitism and clustering along the Pareto front [Zitzler1998]. The technical report was later published [Zitzler1999]. The Strength Pareto Evolutionary Algorithm was developed as a part of Zitzler's PhD thesis [Zitzler1999a]. Zitzler, Laumanns, and Thiele later extended SPEA to address some inefficiencies of the approach, the algorithm was called SPEA2 and was released as a technical report [Zitzler2001] and later published [Zitzler2002]. SPEA2 provides fine-grained fitness assignment, density estimation of the Pareto front, and an archive truncation operator.

Learn More

Zitzler, Laumanns, and Bleuler provide a tutorial on SPEA2 as a book chapter that considers the basics of multiple objective optimization, and the differences from SPEA and the other related Multiple Objective Evolutionary Algorithms [Zitzler2004].

Bibliography

Please Note: This content was automatically generated from the book content and may contain minor differences.