For researchers & practitioners · 15 metaheuristics · continuous & discrete
Find the best solution to your optimization problem — and the algorithm that gets you there, proven by statistics.
Type your objective as a maths formula — continuous, integer or binary — or pick a benchmark, knapsack or TSP, choose from fifteen algorithms, and OptiSim runs seeded, budget-matched experiments on dedicated hardware. You get statistics, significance tests and convergence charts ready for your paper, report or design decision.
Researcher and Lab plans · drag-and-drop workflow builder · results you can export to CSV
The problem
Comparing metaheuristics properly is slow: re-implementing algorithms, matching evaluation budgets, running 30+ seeded trials per algorithm, then doing the statistics. Small differences in setup make results hard to trust or reproduce.
The OptiSim way
One tested engine, the same evaluation budget for every algorithm, a recorded random seed for every run, and a Wilcoxon rank-sum test against the best algorithm — automatically. Share the experiment file and anyone can reproduce your numbers exactly.
Fifteen metaheuristics, continuous and discrete, one fair arena
Continuous problems
Real-valued variables with bounds: 10 benchmark functions (Sphere, Rastrigin, Rosenbrock, Ackley…) or your own formula with constraints.
Integer & binary problems
Whole-number or 0/1 variables in your own formula, plus OneMax, deceptive Trap-5 and the 0/1 knapsack problem.
Permutation problems
Travelling salesman problems — random reproducible instances or your own city coordinates; the best tour comes back as a visiting order.
ACS — Ant Colony
ACOR for continuous problems and Ant Colony System for TSP and other discrete problems (Socha & Dorigo; Dorigo & Gambardella).
GA — Genetic Algorithm
Tournament selection and elitism with SBX/polynomial operators (continuous) or uniform, order crossover and bit-flip/inversion (discrete).
PSO — Particle Swarm
Global-best PSO with decreasing inertia and velocity clamping. Fast, few parameters, a classic baseline.
TS — Tabu Search
Ring neighbourhoods and tabu balls for continuous problems; tabu moves, aspiration and restarts for discrete ones.
DE — Differential Evolution
DE/rand/1/bin (Storn & Price): simple, robust and a standard reference in evolutionary computation.
SA — Simulated Annealing
Metropolis acceptance with exponential cooling over the evaluation budget, for continuous and discrete moves.
CMAES — CMA-ES
Learns the shape of the landscape with a full covariance matrix — state of the art on ill-conditioned continuous problems.
ABC — Artificial Bee Colony
Employed, onlooker and scout bees balance refinement and exploration (Karaboga).
GWO — Grey Wolf Optimizer
The pack follows its three best wolves, shifting from exploration to exploitation over the run (Mirjalili).
WOA — Whale Optimization
Humpback whales encircle the best solution and attack along a bubble-net spiral (Mirjalili & Lewis).
FA — Firefly Algorithm
Fireflies are drawn to brighter ones, with attraction that fades with distance — good at finding several basins (Yang).
CS — Cuckoo Search
Heavy-tailed Lévy flights mix long jumps with local steps; poor nests are abandoned and rebuilt (Yang & Deb).
TLBO — Teaching–Learning
A teacher phase and a learner phase — no algorithm-specific parameters to tune (Rao, Savsani & Vakharia).
JAYA — Jaya
Move towards the best solution and away from the worst; parameter-free and very simple (Rao).
SCA — Sine Cosine Algorithm
Solutions oscillate around the best with sine and cosine steps that shrink over the run (Mirjalili).
Drag-and-drop workflows
Build an experiment visually: drop a problem, the algorithms and a budget on the canvas, connect them and tune every parameter with built-in explanations and valid ranges.
Your formula, safely
Write sum(x**2) + 10*cos(x1*x2), add constraints like x1 + x2 >= 1, set bounds. Formulas are parsed with a strict whitelist — no code runs.
Statistics included
Best, mean, median, std, worst, time per run, feasibility rate, ranks and Wilcoxon p-values — plus mean/median convergence curves.
Never lose a run
Long experiments are checkpointed. If anything restarts, the job resumes and gives exactly the same results as an uninterrupted run.
How it works
Define
Pick one of 10 benchmarks or type your objective, dimension, bounds and constraints.
Configure
Drag the algorithms into a workflow (or use the form), tune their parameters, set runs, evaluation budget and seed.
Run
Your experiment is queued on our compute node; watch live progress and convergence.
Publish
Download CSV and the experiment file, copy tables, export charts for your paper or report.
Proof point: on the 2-variable constrained test problem (x1−2)² + (x2−1)² with x1² ≤ x2 and x1 + x2 ≤ 2, all four algorithms find the known optimum f = 1 at x = (1, 1) in 10 out of 10 runs.
Dedicated compute
Runs on our own compute machines
Your experiments do not share a noisy cloud VM. They run on OptiSim's own dedicated machines in Malaysia, so run times are consistent and comparable between algorithms. Run on any available machine, or choose a node group — each one lists its CPU cores, memory and GPU before you submit.
Each experiment gets its own CPU core with single-threaded maths, which keeps time-per-run statistics fair. Every run is checkpointed: after a power cut or restart the job resumes and gives exactly the same results.
How your experiments run
| Machines | Dedicated computers owned and run by OptiSim — no shared cloud VMs |
|---|---|
| Your choice | Run on any available machine, or pick a node group; each group shows its CPU cores, memory and GPU |
| Fair timing | One CPU core per experiment with single-threaded maths |
| Reliability | Every run is checkpointed and resumes after a restart with identical results |
| Location | Malaysia — your data stays in-country |
Plans
Both plans include every algorithm and feature. They differ in problem size, experiment size, monthly compute and team features. Prices are quoted on request — tell us about your research and we will recommend a plan.
Researcher
RecommendedContact us
For individual researchers, PhD students and practitioners running their own studies.
| Decision variables (D) | up to 100 dimensions per problem |
|---|---|
| Independent runs | up to 51 per algorithm — enough for the 51-run CEC protocol |
| Evaluations per run | up to 300,000 budget per run — 10,000 × D for D up to 30 |
| Monthly compute | 200 M evaluations ≈ 3 full studies a month* |
Every feature, plus:
- Personal experiment workspace
- Share any experiment with your lab when you join one
Lab
Contact us
For research groups and teams who run large studies and want one shared workspace.
| Decision variables (D) | up to 1,000 dimensions per problem |
|---|---|
| Independent runs | up to 100 per algorithm — enough for the 51-run CEC protocol |
| Evaluations per run | up to 3 M budget per run — 10,000 × D for D up to 300 |
| Monthly compute | 2 B evaluations ≈ 32 full studies a month* |
Everything in Researcher, plus:
- Lab workspace: members share experiments with the whole group
- Lab Manager sees every lab experiment, members and monthly lab usage
- Lab Manager adds and removes members by email
- We set up your lab and appoint your Lab Manager
Included in every plan
- All 15 algorithms — ACS, GA, PSO, TS, DE, SA, CMA-ES, ABC, GWO, WOA, FA, CS, TLBO, Jaya, SCA — with every parameter tunable
- Continuous, integer, binary and permutation (TSP) problems
- Drag-and-drop workflow builder and a classic form editor
- 10 benchmark functions plus your own formulas with bounds and constraints
- Private custom problems implemented for you on request (paid service, quoted per problem)
- Identical evaluation budget and recorded seeds for every algorithm
- Best, mean, median, std, worst, time per run, feasibility rate and ranks
- Wilcoxon rank-sum tests against the best algorithm
- Mean and median convergence charts
- CSV export and reusable experiment files (also open in the desktop app)
- Runs on dedicated compute machines in Malaysia, checkpointed against restarts
- Optional two-factor authentication and a full activity log
- Email support
* An experiment uses algorithms × runs × evaluations per run function evaluations, shown before you submit. A “full study” here means 4 algorithms × 51 runs × 300,000 evaluations (61.2 M). Usage resets on the 1st of each month (Malaysia time); failed experiments are not counted.
Frequently asked questions
What is OptiSim?
OptiSim is an online optimization simulator for continuous and discrete problems. It runs fifteen metaheuristics — Ant Colony, Genetic Algorithm, Particle Swarm, Tabu Search, Differential Evolution, Simulated Annealing, CMA-ES, Artificial Bee Colony, Grey Wolf Optimizer, Whale Optimization, Firefly, Cuckoo Search, Teaching–Learning-Based Optimization, Jaya and Sine Cosine — on the same evaluation budget and compares them with statistics and convergence charts.
Can I solve discrete and combinatorial problems?
Yes. Variables can be continuous, integer, binary (0/1) or a permutation. GA, Simulated Annealing, Tabu Search and Ant Colony have native discrete versions; the other eleven search continuous keys that are decoded by rounding, thresholding or sorting (random keys). Built-in discrete problems include OneMax, Trap-5, the 0/1 knapsack and the travelling salesman problem, and you can enter your own TSP cities.
Can I optimize my own objective function?
Yes. Type the objective as a maths formula such as sum(x**2) + 10*cos(x1*x2), set the dimension and bounds, and add inequality or equality constraints. Formulas are parsed with a strict whitelist, so no code is executed.
My problem does not fit a formula — can you help?
Yes. If your objective needs a simulation, measured data, a lookup table or several steps of code, send us a private problem request from your account (or the contact form). Custom problems are a paid service: we reply with a quote (a one-time fee that depends on the work), implement and test it with you, and once paid it appears as a private problem in your account only — no other user can see or run it — and works with every algorithm, statistic and report.
Which benchmark problems are included?
Ten continuous functions — Sphere, Rosenbrock, Rastrigin, Ackley, Griewank, Schwefel, Levy, Zakharov, Styblinski–Tang and High-Conditioned Elliptic — plus four discrete problems: OneMax, Trap-5, 0/1 knapsack and TSP (random instances generated from a seed, so they are reproducible).
Which statistics do I get?
Best, mean, median, standard deviation and worst objective value, time per run, feasibility rate, ranks, Wilcoxon rank-sum p-values against the best algorithm, and mean/median convergence curves — exportable to CSV.
Are the results reproducible?
Yes. Every run uses a recorded random seed and the same evaluation budget for every algorithm. Share the experiment file and anyone can reproduce your numbers exactly.
Where do my experiments run?
On OptiSim's own dedicated compute machines in Malaysia, not on shared cloud machines. You can let the first free machine take it or choose a node group, and each group lists its CPU cores, memory and GPU. Long experiments are checkpointed and resume automatically after a restart.
Which plans are available?
Two plans: Researcher for individual researchers and practitioners, and Lab for research groups that need higher dimensions, more runs and larger evaluation budgets. Contact us for pricing and a demo.
Security you can rely on
- Argon2id password hashing, account lockout and rate limiting
- Optional two-factor authentication for every account (authenticator app or email code, with recovery codes)
- Role-based permissions enforced on every request; complete audit log
- Encrypted connections; your experiments are private unless you share them with your lab
- Personal data handled in line with Malaysia's PDPA 2010
Request a demo or a Lab plan
Sales & partnerships: marketing@alaganu.com
Support: committee@alaganu.com