Particle Swarm Optimization PSO: How Particles Update Positions and Velocities with a Flocking Analogy | AI Research

Particle Swarm Optimization (PSO): How Particles Update Positions and Velocities (with a Flocking Analogy)

Verified Sources

Nov 4, 2026

Particle Swarm Optimization keyword(PSO) keyworduses a swarm of particles to search for an optimum in a space of decision variables. Each particle “moves” through the search space by updating its keywordvelocity and keywordposition. The motion is guided by two memories: the particle’s own best position found so far (its keywordpbest) and the best position found by the entire swarm (its keywordgbest). In common form, PSO balances three effects: (i) inertia (continue moving as before), (ii) attraction to pbest (learn from self), and (iii) attraction to gbest (learn from the group).2

A simple birds-flocking analogy:

Imagine each particle is a bird flying over terrain.
The bird’s current keywordvelocity is how fast and in which direction it’s currently flying.
The bird also “remembers” the best spot it has ever found (pbest).
Birds can observe the best spot the flock has found so far (gbest).
At each moment, the bird keeps some of its prior motion (inertia), then adjusts direction toward its own best and the flock’s best—while some randomness helps exploration and prevents the flock from getting stuck too early.2

Mermaid mental model:

Particle swarm optimization (PSO) overview and update equations — https://en.wikipedia.org/wiki/Particle_swarm_optimization - Defines velocity/position updates, inertia weight, and pbest/gbest concepts. ↩ ↩²
Standard PSO algorithm description and parameter roles — https://www.geeksforgeeks.org/particle-swarm-optimization/ - Explains PSO steps and interprets inertia/cognitive/social terms in the iteration process. ↩ ↩²

Particle Swarm Optimization (PSO) - How it Works

Core PSO update rules: positions and velocities

Let particle $i$ at iteration $t$ have:

position vector $x_i(t)$ (candidate solution),
velocity vector $v_i(t)$ (movement),
personal best position $pbest_i$ ,
global best position $gbest$ .

A widely used PSO update uses these equations (vector form):2

Velocity update

v_i(t+1)=w\,v_i(t)+c_1\,r_1\,(pbest_i-x_i(t))+c_2\,r_2\,(gbest-x_i(t))

Position update

x_i(t+1)=x_i(t)+v_i(t+1)

Where:

$w$ is the inertia weight controlling how much the particle retains its current velocity.2
$c_1$ and $c_2$ are acceleration coefficients controlling how strongly a particle is pulled toward $pbest_i$ and $gbest$ respectively.2
$r_1$ and $r_2$ are random numbers in $[0,1]$ , typically sampled independently at each iteration (and usually per dimension), adding stochastic exploration.2

Key interpretation (ties to birds flocking)

Inertia term $w\,v_i(t)$ : “Keep some of your previous flying direction and speed.”2
Cognitive term $c_1\,r_1\,(pbest_i-x_i(t))$ : “Turn toward where you personally did best.”2
Social term $c_2\,r_2\,(gbest-x_i(t))$ : “Also adjust toward where the flock did best.”2

Particle swarm optimization (PSO) overview and update equations — https://en.wikipedia.org/wiki/Particle_swarm_optimization - Defines velocity/position updates, inertia weight, and pbest/gbest concepts. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷
Standard PSO algorithm description and parameter roles — https://www.geeksforgeeks.org/particle-swarm-optimization/ - Explains PSO steps and interprets inertia/cognitive/social terms in the iteration process. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

One PSO iteration (particle update loop)

1
Step 1
Pick $N$ particles with random positions $x_i(0)$ and velocities $v_i(0)$ . Evaluate each particle’s fitness at its starting position.
2
Step 2
For each particle, set $pbest_i=x_i(0)$ . Track $gbest$ as the best $pbest_i$ across the swarm.
3
Step 3
Compute $v_i(t+1)=w\,v_i(t)+c_1\,r_1\,(pbest_i-x_i(t))+c_2\,r_2\,(gbest-x_i(t))$ using new random $r_1,r_2$ each iteration.
4
Step 4
Move the particle: $x_i(t+1)=x_i(t)+v_i(t+1)$ .
5
Step 5
Compute the objective/fitness value at $x_i(t+1)$ .
6
Step 6
If the new position improves the particle, update $pbest_i=x_i(t+1)$ . Then update $gbest$ to the best $pbest_i$ over all particles.
7
Step 7
Stop after a fixed number of iterations, or when improvement becomes small, or when a target fitness is reached.

How each PSO term influences motion (conceptual contributions)

These are qualitative weights: higher magnitude typically causes stronger pull in that direction.

Pro Tip: Why the random numbers matter

The $r_1$ and $r_2$ terms make the cognitive and social pulls vary every iteration, which helps the swarm explore different regions instead of collapsing into a single trajectory too early.2

Particle swarm optimization (PSO) overview and update equations — https://en.wikipedia.org/wiki/Particle_swarm_optimization - Defines velocity/position updates, inertia weight, and pbest/gbest concepts. ↩
Standard PSO algorithm description and parameter roles — https://www.geeksforgeeks.org/particle-swarm-optimization/ - Explains PSO steps and interprets inertia/cognitive/social terms in the iteration process. ↩

Common implementation pitfall: update ordering

Be consistent about whether you use $v_i(t+1)$ immediately for $x_i(t+1)$ . The standard form shown here updates velocity first, then position using the new velocity: $x_i(t+1)=x_i(t)+v_i(t+1)$ . Mixing $v_i(t)$ vs $v_i(t+1)$ can change behavior substantially.2

Particle swarm optimization (PSO) overview and update equations — https://en.wikipedia.org/wiki/Particle_swarm_optimization - Defines velocity/position updates, inertia weight, and pbest/gbest concepts. ↩
Standard PSO algorithm description and parameter roles — https://www.geeksforgeeks.org/particle-swarm-optimization/ - Explains PSO steps and interprets inertia/cognitive/social terms in the iteration process. ↩

A tiny numeric walk-through (1D) to make the updates tangible

Assume a 1D objective, so everything is scalars. At iteration $t$ for particle $i$ :

current position: $x_i(t)=2$
current velocity: $v_i(t)=0.5$
personal best: $pbest_i=5$
global best: $gbest=4$
parameters: $w=0.7$ , $c_1=1.5$ , $c_2=1.5$
random draws: $r_1=0.3$ , $r_2=0.8$

Velocity update

v_i(t+1)=0.7(0.5)+1.5(0.3)(5-2)+1.5(0.8)(4-2)

v_i(t+1)=0.35+1.5(0.3)(3)+1.5(0.8)(2)

v_i(t+1)=0.35+1.35+2.4=4.10

Position update

x_i(t+1)=x_i(t)+v_i(t+1)=2+4.10=6.10

Interpretation in flocking terms: the particle had inertia (kept moving in its prior direction), but strong social pull toward $gbest$ (because $gbest-x_i(t)$ is large and $r_2$ was large) made it turn more aggressively than the cognitive term.2

Particle swarm optimization (PSO) overview and update equations — https://en.wikipedia.org/wiki/Particle_swarm_optimization - Defines velocity/position updates, inertia weight, and pbest/gbest concepts. ↩
Standard PSO algorithm description and parameter roles — https://www.geeksforgeeks.org/particle-swarm-optimization/ - Explains PSO steps and interprets inertia/cognitive/social terms in the iteration process. ↩

Common PSO concepts (mapped to the flocking analogy)

PSO search cycle (repeated iterations)

Initialize

t = 0

Random positions $x_i(0)$ , velocities $v_i(0)$ ; set $pbest_i$ and $gbest$ ."

Move

t → t+1 (repeat)

Update $v_i(t+1)$ using inertia + cognitive + social; then set $x_i(t+1)=x_i(t)+v_i(t+1)$ ."

Remember

After moving

Evaluate fitness; update $pbest_i$ and then $gbest$ ."

Termination

Stop

Stop at max iterations or when improvements fall below a threshold."

PSO Quick Recall

1 / 4

Question · Term

What are the two quantities each PSO particle maintains?

Click to reveal

Answer · Definition

A position $x_i$ (candidate solution) and a velocity $v_i$ (how it moves).

Knowledge Check

Question 1 of 3

Q1Single choice

In the standard PSO update, which term pulls a particle toward its own best position?

$w\,v_i(t)$ (inertia)

$c_1\,r_1\,(pbest_i-x_i(t))$ (cognitive term)

$c_2\,r_2\,(gbest-x_i(t))$ (social term)

$x_i(t+1)=x_i(t)+v_i(t)$ (position update)

Explore Related Topics

Why Use Ensemble Methods? And How Do They Work?

Best Way to Approach a Game-Playing Problem: Linear vs Heuristic vs Random vs Optimal

CSMA/CA (Collision Avoidance) — Principles, Timing, and Decision Flow

Browse all research articles

Particle Swarm Optimization (PSO): How Particles Update Positions and Velocities (with a Flocking Analogy)

Footnotes

Particle Swarm Optimization (PSO) - How it Works

Core PSO update rules: positions and velocities

Key interpretation (ties to birds flocking)

Footnotes

One PSO iteration (particle update loop)

How each PSO term influences motion (conceptual contributions)

Pro Tip: Why the random numbers matter

Footnotes

Common implementation pitfall: update ordering

Footnotes

A tiny numeric walk-through (1D) to make the updates tangible

Footnotes

Common PSO concepts (mapped to the flocking analogy)

PSO search cycle (repeated iterations)

Initialize

Move

Remember

Termination

PSO Quick Recall

What are the two quantities each PSO particle maintains?

Knowledge Check

Explore Related Topics