Simulation Optimization App

The Simopt app (simulation optimization) searches for the model inputs that give the best response. You define decision variables and bounds, pick an objective to minimize or maximize, choose a search algorithm, and the app runs the model many times to find a good solution.

You will need: Java 21 and a model bundle. This guide uses an LK (s,S) inventory optimization model from the KSL Book Examples bundle. New here? Read Common UI & concepts, then the Experiment guide.

What you'll be able to do

• Choose an objective response and decision variables with bounds.
• Pick a search algorithm (e.g. simulated annealing).
• Run the optimization and read the best solution it found.

---

1. At a glance

Unlike the tab-based apps, Simopt is a stepper: you move through Model → Problem → Constraints → Algorithm → Run Setup → Execute → Results, completing one step at a time.

Use this app when…	Use a sibling app when…
You want the computer to find the best inputs.	You already have configs to compare → Scenario app
The input space is too large to enumerate.	You want to measure effects over a grid → Experiment app

---

2. Before you begin

Load the model from a bundle (see Common UI → Models and bundles). The model must expose the inputs you want to optimize as controls and the objective as a response.

---

3. A guided tour of the window

1. Menu bar — File, Bundles, Help.
2. Stepper rail — the seven steps; a ✓ marks completed steps. Click a completed step to revisit it.
3. Step footer — Back / Next to move between steps, with an Edited/Saved badge.
4. Console drawer — the optimization log.

---

4. Tutorial — minimize inventory cost

Step 1 — Model

Select the LK Inventory (optimization) model. The step shows its controls (candidate decision variables) and responses (candidate objectives).

Step 2 — Problem

Set the objective: response TotalCost, direction Minimize. Then Add input… for each decision variable with its bounds:

Decision variable	Lower	Upper	Granularity
Inventory.reorderPoint	1	60	1
Inventory.orderQuantity	1	100	1

Granularity 1 keeps the search on whole units. The indifference zone (Δ) is the smallest objective difference you consider meaningful (leave 0 for full precision).

Step 3 — Constraints (optional)

Add bounds or linear constraints relating the variables/responses, or skip straight through — this step auto-completes when the problem is defined.

Step 4 — Algorithm

Choose a search algorithm — Simulated Annealing, Stochastic Hill Climbing, Cross-Entropy, or R-SPLINE — and set its parameters (iterations, replications per evaluation, cooling schedule, …).

Step 5 — Run Setup

Set the Evaluation settings (result caches, snapshot frequency, feasibility, precision, parallel workers) and the Tracking & trace options. The iteration budget itself lives on the Algorithm step, and the output folder is named by the Analysis name field in the toolbar.

Step 6 — Execute

Click Optimize. The step shows live progress — current best objective and iteration count — and the console logs each evaluation.

Reading the results

Step 7, Results, reports the best solution. Below is a genuine run of simulated annealing on this problem (40 iterations, 20 replications per evaluation) — the same solver report the app produces.

The search ran 31 iterations and made 132 evaluations (2,640 replications). It improved from the initial guess to the best solution found:

	Reorder Point	Order Quantity	Avg Total Cost (95% half-width)
Initial solution	33	14	124.65 (± 1.13)
Best found	14	49	122.31 (± 1.96)

How to read it. Simulated annealing explored combinations of reorder point and order quantity, keeping the lowest-cost solution: a reorder point of 14 and order quantity of 49, giving an estimated total cost of 122.31 per period. Because the objective is itself a simulation estimate, it comes with a confidence half-width — solutions whose intervals overlap are not clearly different, which is what the indifference zone setting is for.

---

5. Reference — every step explained

Step	What it's for
Model	Choose the model; view controls and responses.
Problem	Objective response + direction; decision variables with bounds and granularity.
Constraints	Optional variable/response bounds and linear constraints.
Algorithm	Pick the search algorithm, its parameters, and the iteration budget.
Run Setup	Evaluation settings (caches, precision, parallel workers) and tracking/trace.
Execute	Run the optimization; watch live progress and the current best.
Results	The best solution found, with downloadable artifacts.

---

6. Common tasks

Task	How
Revisit an earlier step	Click it on the stepper rail (if completed)
Constrain the search	Add bounds/constraints on the Constraints step
Make the search longer/shorter	Adjust iterations and replications on Algorithm / Run Setup
Treat near-ties as equal	Raise the indifference zone (Δ) on the Problem step
Save / reopen the study	File → Save / Open (a .toml document)

---

7. Troubleshooting & gotchas

Symptom	Cause	Fix
A later step is locked	An earlier step is incomplete.	Complete the objective + at least one decision variable.
Optimization is very slow	Many evaluations × many replications.	Lower the iteration count or replications per evaluation.
Best solution looks no better than the start	Too few iterations, or a flat/noisy objective.	Run longer, or raise replications to reduce noise.
Validation errors block Optimize	The problem is incomplete (e.g. no objective response chosen).	Open the details on the Execute step; set the objective on the Problem step.

---

8. See also

• Common UI & concepts · Experiment app · Scenario app · Results app
• KSL Book — simulation optimization.