How does optimization of simulation experiments work?

General (quick video)

The optimization of simulation experiments in Simulation is based on experiments with factor variation. In an experiment with factor variation, you can vary the attribute values for the objects included in the simulation. Depending on the configuration, many scenarios are created. You can specify an optimizing function to find the optimal configuration of processes and resources without having to simulate all kinds of possible scenarios.

Calculate the target function

The target function for the optimization is calculated by multiplying each result KPI with the specified weighting and then adding the weighted results.

Example

The following weightings are specified for the results:

Weighting

Result

Value

0,5

Throughput time (avg.)

3.600

-1.000

Process folders in dynamic wait state

10

0

Processing time (avg.)

300

Calculation:

0.5 x 3,600 - 1,000 x 10 + 0 x 300 = -8,200

The value for this target function is -8.200 and is displayed in the output file in the Target column.

Calculate the number of simulation runs

The goal of optimization is finding the optimal result with a given number of simulation runs without checking all possible scenarios. The number of possible configurations results from the number of specified factors, their upper and lower limits, and the step width. The number of simulations runs to be performed is determined by the following table:

Number of configurations and simulation runs

Number of possible configurations

Number of simulation runs

101 to 1,000

100

1,001 to 5,000

200

5,001 to 100,000

200 + 20 for each additional 5,000 configurations over 5,001

100,001 to 1,000,000

380 + 20 for each additional 10,000 configurations over 100,001

1,000,001 to 100,000,000

560 + 20 for each additional 1,000,000 configurations over 1,000,001

From 100,000,001

2,540 + 20 for each additional 10,000,000 configurations over 10,000,001

The optimization becomes active only from 101 configurations.

Example

Two factors are specified, with the first factor varying between 1 and 100 and the second between 1 and 120. The step width is 1. The result is 100 x 120 = 12,000 possible configurations. According to the table, 240 simulation runs result.