Particle Swarm Optimization (PSO) using C#

Introduction

In the realm of optimization algorithms, Particle Swarm Optimization (PSO) stands out as an efficient and intuitive technique inspired by social behavior. Developed in the 1990s by Eberhart and Kennedy, PSO has gained popularity in various fields, including engineering, computer science, and economics. By simulating the movement of a swarm of particles, PSO searches for optimal solutions in complex problem spaces.

In this post, we will delve into the essence of PSO, employing a relatable metaphor to facilitate understanding and present a basic implementation in C#. If you're interested, on my blog, I often write about artificial intelligence, including particle swarm optimization, genetic algorithms, neural networks, and other biologically inspired algorithms.

Read more on my blog or watch me on YouTube!

PSO through the flock of birds Metaphor

To grasp the concept of PSO, let's imagine a flock of birds searching for the most favorable location to roost for the night. Each bird represents a potential solution to the optimization problem. The goal is to find the optimal location analogous to the solution we seek.

Initially, the birds scatter randomly across the sky. However, they are not entirely clueless about their surroundings. Each bird has limited vision and can only perceive its distance and relative position to a few nearby birds. Similarly, in PSO, each particle represents a potential solution, and its position in the problem space defines its fitness.

Now, as the birds start to explore, they remember the position of the best roost they have discovered so far. If a bird finds a better roost, it adjusts its flight path to move closer to that location. In PSO, each particle remembers its own best position (pBest) and the global best position (gBest) discovered by any particle within the swarm.

Gradually, the flock becomes more synchronized. As birds observe others moving toward the gBest location, they tend to follow suit, creating a cooperative and coordinated movement. Similarly, in PSO, particles adjust their positions towards the global best position, promoting information sharing and convergence towards the optimal solution.

Implementation in C#

Let's now explore a basic implementation of PSO in C#. We'll consider a simple problem of finding the minimum of a function within a specified range. The objective function we're minimizing is x^2 + y^2, so our particles have 2 dimensions: x and y. The optimal solution where x and y are at their minimum is x = 0 and y = 0, so we should expect that the algorithm finds values close to zero.

Source code

public class Program
{
    private const int NumParticles = 30;
    private const int NumDimensions = 2;
    private const int MaxIterations = 100;

    private static double ProblemFunction(double x, double y)
    {
        return Math.Pow(x, 2) + Math.Pow(y, 2);
    }

    public static double[] ParticleSwarmOptimization()
    {
        // Initialize the particles and velocities
        List<double[]> particles = new List<double[]>();
        List<double[]> velocities = new List<double[]>();

        Random random = new Random();

        for (int i = 0; i < NumParticles; i++)
        {
            double[] particle = new double[NumDimensions];
            double[] velocity = new double[NumDimensions];

            for (int j = 0; j < NumDimensions; j++)
            {
                particle[j] = random.NextDouble() * 10;
                velocity[j] = random.NextDouble();
            }

            particles.Add(particle);
            velocities.Add(velocity);
        }

        // Initialize the best positions
        List<double[]> pBest = new List<double[]>();
        double[] gBest = particles[0];

        foreach (var particle in particles)
        {
            pBest.Add((double[])particle.Clone());
        }

        // Perform iterations
        for (int iteration = 0; iteration < MaxIterations; iteration++)
        {
            for (int i = 0; i < NumParticles; i++)
            {
                double[] particle = particles[i];
                double[] velocity = velocities[i];
                double[] pBestParticle = pBest[i];

                for (int j = 0; j < NumDimensions; j++)
                {
                    velocity[j] = velocity[j] + random.NextDouble() * (pBestParticle[j] - particle[j]) + random.NextDouble() * (gBest[j] - particle[j]);
                    particle[j] = particle[j] + velocity[j];
                }

                // Update pBest and gBest
                if (ProblemFunction(particle[0], particle[1]) < ProblemFunction(pBestParticle[0], pBestParticle[1]))
                {
                    pBest[i] = (double[])particle.Clone();
                }

                if (ProblemFunction(particle[0], particle[1]) < ProblemFunction(gBest[0], gBest[1]))
                {
                    gBest = (double[])particle.Clone();
                }
            }
        }

        return gBest;
    }

    static void Main(string[] args)
    {
        double[] solution = ParticleSwarmOptimization();

        Console.WriteLine("Optimal solution: (" + solution[0] + ", " + solution[1] + ")");
    }
}

Summary

Particle Swarm Optimization (PSO) provides an elegant approach to solve complex optimization problems by simulating the collective behavior of particles in a swarm. By sharing information and adapting their positions, particles converge toward an optimal solution. In this article, we introduced the fundamentals of PSO using the metaphor of a flock of birds searching for the best roosting location. We also presented a basic implementation of PSO in C#, enabling you to explore its application in various domains. PSO's versatility and simplicity make it a valuable tool in tackling optimization challenges.