Introduction
Genetic algorithms (GA) are adaptive heuristic search algorithms based on the evolutionary ideas of natural selection and genetics. GA’s operate on a population of potential solutions applying the principle of survival of the fittest to produce better and better approximations to a solution. At each generation, a new set of approximations is created by the process of selecting individuals according to their level of fitness in the problem domain and breeding them together using operators borrowed from natural genetics. This process leads to the evolution of populations of individuals that are better suited to their environment than the individuals that they were created from, just as in natural adaptation.
Genetic algorithms find application in bioinformatics, phylogenetics, computational science, engineering, economics, chemistry, manufacturing, mathematics, physics and other fields. They can be used to generate high-quality solutions to optimization and search problems by relying on biologically inspired operators such as mutation, crossover and selection. GA’s are well suited for problems where the search space is discrete, evolvable, and very large or infinite. Typical GA representation uses binary strings of 0s and 1s, but other encodings are also possible. This article discusses the process of genetic algorithms in detail and provides references to relevant research papers applying GA’s to solve various real-world problems.
Representation
The first step in applying a genetic algorithm is to choose an appropriate representation for potential solutions to the problem. Typically, a chromosome or individual represents a solution in the form of a bit string or other comparable data structure. For some problems, it may make sense to encode potential solutions as real-valued instead of binary. The length of the chromosome and the mapping between chromosome and phenotype determine how many variations are possible in candidate solutions to the problem. Binary digits can encode candidate solutions for problems involving discrete or real-valued parameters. Real-valued encoding maps chromosomes directly to corresponding phenotypes without the need for translation. Whether to use binary or real-valued encoding depends on the problem domain.
For example, in a classic traveling salesman problem where the goal is to find the shortest route between cities, chromosome encoding could represent an ordering of cities as a string of integers from 1 to the number of cities. A candidate solution of the tour “2-4-3-1-5” could be encoded as the chromosome “24315”. In some cases, it may be more efficient to encode cities directly as real values rather than integers. The choice of encoding significantly impacts how well variations can explore the search space. Well-chosen encoding helps genetic algorithms work better for a given problem by matching the representation to the structure of the solutions.
Fitness Function
A fitness function is defined to evaluate the goodness of each candidate solution represented by chromosomes. This measure drives the evolutionary process by enabling selection to be elitist—solutions with a higher fitness rating will be more likely to reproduce. The fitness function should capture the essence of the optimization problem, representing the goal or objective to optimize. It is a key component that shapes the search space.
In a minimization problem, solutions with lower fitness values are better. While in maximization problems, solutions with higher fitness are preferable. Fitness function values can range from 0 to 1 or be unbounded. Scaled to similar ranges often helps genetic algorithms converge faster. Care must be taken to properly formulate the fitness function so that it effectively guides the search towards optimal solutions and does not mislead the evolutionary process. The choice of encoding and fitness function together define the search space and fitness landscape.
For the traveling salesman problem, a simple fitness function could compute the total distance traveled in a candidate route. Shorter routes will correspond to higher fitness. In engineering design optimization tasks, objectives like minimizing cost, weight or maximizing efficiency can serve as the fitness metric. Multimodal problems involving conflicting objectives may require aggregating multiple goals into a scalar fitness value. Defining the right fitness function is crucial for Genetic Algorithms to work well on any given problem.
Genetic Operators
After calculating the fitness of initial chromosomes, genetic operators are applied during evolution to produce the next generation population. The main genetic operators used are:
Selection: It selects parent chromosomes from the population for later reproduction. Fitness-proportionate or elite selection favor fitter chromosomes with better chance to be selected. Selection allows survival of the fittest individuals.
Crossover: It mates two parent chromosomes and randomly exchanges parts of them to form new offspring with hope that they may be even better adapted. Single or multi-point crossover works on binary strings, while arithmetical crossover blends real-valued parents. Crossover enhances global exploration ability.
Mutation: It randomly flips some of the bits in chromosomes with low probability of mutation rate. Mutation avoids getting stuck in local optima and maintains diversity. It provides local exploration to search unvisited areas.
The selected parents undergo genetic operations to breed new offspring replacing the old population for next generation. Generations evolve through iterative application of selection, crossover and mutation. Eventually, after many generations, chromosome with highest fitness emerges as near-optimal solution for the problem. Parameters like population size, crossover rate and mutation rate need tuning to achieve faster convergence.
Algorithm Workflow
The standard genetic algorithm follows these steps in each generation of evolution:
Initialize a random population of chromosomes (candidate solutions).
Evaluate the fitness of each chromosome using the fitness function.
Repeat these steps until some criterion is met (solution is found, or max generations exceeded):
3.1 Select parents for reproduction from the population proportionally based on their fitness.
3.2 Breed new offspring between parent chromosomes by applying genetic operators
like crossover and mutation.
3.3 Evaluate the fitness of new offspring chromosomes.
3.4 Replace a part of the old population with new offspring population.
Once criteria is met, terminate and return the best solution individual.
This evolutionary cycle continues, passing traits to offspring over many generations, until optimal or near-optimal solutions emerge. Factors like population size, selection method, crossover/mutation rates affect performance and convergence behavior. Genetic algorithms can be easily parallelized since fitness evaluation and genetic operations on each chromosome are independent of others. Overall genetic algorithms are well-suited stochastic search and optimization tools for complex problems.
Applications
Genetic algorithms find diverse real-world applications due to their ability to evolute optimal/near-optimal solutions with minimal assumptions about the problem being optimized. Some well-known applications areas include:
Scheduling: For optimization of production, workforce and resource scheduling problems in manufacturing and services. (Ramesh et al., 2021)
Engineering Design: To evolve optimal designs automatically for structures, integrated circuits, antennas etc. subject to constraints. (Chang et al., 2013)
Data Mining: Feature selection, clustering, classification, association rule mining benefit from GA’s search capabilities. (Mitra et al., 2002)
Bioinformatics: Sequence alignment, gene selection, microarray data analysis and phylogenetics. (Jiang et al., 2019)
Cyber Security: For intrusion detection system optimization, cryptanalysis, digital forensics. (Goh et al., 2022)
Finance: Portfolio optimization, option pricing, stock prediction and trading strategies. (Yu et al., 2018)
Traveling Salesman Problem: Evolving optimal traveling routes meeting constraints. (Cho and Chang, 2020)
Vehicle Routing Problem: Optimization of goods delivery vehicle routing. (Zhou et al., 2018)
Genetic algorithms’ wide applicability arises from their capacity to find good quality solutions when traditional methods fail due to nonlinearity, discontinuities in the solution space or unavailability of derivatives etc. Hybridization with other metaheuristics has enhanced performance on challenging domains.
Conclusion
Genetic algorithms provide a simple yet powerful method for search and optimization problems based on the natural evolution process. With a population of candidate solutions evolving through selection, crossover and mutation, GA’s are well-suited for problems with huge, non-linear search spaces where traditional techniques fail. This paper described the mechanics of genetic algorithms and their essential components like representation, fitness function and genetic operators involved in the evolutionary process. It also provided examples of diverse application domains benefited by genetic algorithms’ optimization capabilities. Overall genetic algorithms constitute a robust, flexible and scalable methodology for both theoretical studies and practical problem solving.
Andrew Stroud