Page 672 - ExtendSim User Guide
P. 672

646 Analysis
Sensitivity analysis
dence level is the probability that the true mean will be located within the range. Typical confi- dence levels are 90%, 95%, and 99%. Notice that, at higher levels of probability, the interval gets wider.
The Statistics and Mean & Variance blocks (Value library) and the Cost Stats block (Item library) provide the option to specify a confidence level in their dialogs. For example, the Sta- tistics block not only summarizes and reports statistical information, but also calculates confi- dence intervals for the information given various levels of confidence.
To generate a confidence interval where each sample is the result from a single simulation run, in the Run > Simulation Setup > Random Numbers tab do one of the following:
• Randomize the seed by entering a blank or zero random seed. • Or, select Continue random number sequence in the popup.
• Or, select Use Database table __Seed for values in the popup.
To obtain sufficient sample data to determine the confidence interval, multiple observations of each statistic must be made and appended to the table of data. This is shown in the “Run for CI” model located in the \Examples\Tips\Modeling Tips folder. The model runs repeatedly based on the Mean & Variance’s Options tab settings Calculate for multiple simulations and Replicate until relative error is <= 0.01.
Sensitivity analysis
Sensitivity analysis allows you to conduct controlled experiments to explore how much of an impact a particular parameter has on model results. ExtendSim’s sensitivity analysis features make it easy and convenient to specify a parameter to investigate and settings to use for the analysis.
You use sensitivity analysis to investigate the effect of changing one or more parameters upon an area of interest. Sensitivity analysis works with all numeric parameter fields. It also works with clones of those numeric items. You can add sensitivity to as many dialog values as you like. However, it is recommended that you only vary one or two dialog values at a time so as not to confuse the analysis. Once a parameter has been sensitized, specify a multiple number of runs and run the simulation.
The resulting values for the area of interest are usually plotted. Although you can use any of the standard plotters from the Plotter library (such as the Plotter I/O or the Plotter Discrete Event), it is more common to use a MultiSim plotter (to show up to four runs at a time in one plot window) or an Error Bars plotter (to show the mean and standard deviation of the parame- ters over the count of runs). The results of varying the parameter value over the selected set- tings will be displayed as the simulation is run multiple times.
Parameters are sensitized by right-clicking or by using the Edit > Sensitize Parameter com- mand. The Edit > Open Sensitized Blocks command shows all the dialogs for blocks that have sensitivity settings, even if sensitivity analysis is not enabled. This is convenient if you have entered sensitivity settings for many parameters in a large model.
☞ To utilize sensitized parameters, the number of simulation runs must be greater than one. Steps for using sensitivity analysis
For this example, use the Reservoir 1 model discussed in the Tutorial module. Open the Reservoir 1 model (located in the \Examples\Tutorials folder)
How To

   670   671   672   673   674