sensitivity analysis

[adjie]

Dear All,

i tried to do sensitivity analysis on my rainfall data. i did several times for the iterations of; 50, 100, 500,600 and 1000. when i did the 50 iteration, the rainfall data is ok. but when i did the 100 iterations, the data in 90th looks invalid. this problem also occur when i did 100, 500, 600 and 1000 iterations. at the 90th, the rainfall become negative in those three iterations. this situation is only at the 90th of iterations. so, i can say that it is once out of 100, 500, 600 and 1000 times. do you think that it is ok? if it is a problem, what kinds of problems are they?

thanks a lot!

[Administrator]

If rainfall is negative, then you need to modify the model to make sure this does not happen.

There are a couple of threads that might help,
http://www.ventanasystems.co.uk/forum/v ... ive#p15408
http://www.ventanasystems.co.uk/forum/v ... ock#p14605

If you can upload the model it will be easier to help you. Or just post the equations for rainfall and we might be able to help.

Tony.

[adjie]

Dear All,

i try to make a simple model with just only two variable (rainfall and noise seed),

sample.mdl

the simple model of rainfall

(1.19 KiB) Downloaded 247 times

. i use formula of RANDOM WEIBULL( 0.001, 0.023625, 0.962, 0, 1, 0). when i simulate the model, it looks good. But then, i do the sensitivity analysis with the parameter of noise seed from 1 to 10.000. after 1000 iterations, the model start giving the errors. do you have any ideas about any mistakes on the model?

Thanks,

[Administrator]

What errors are you getting?

[Administrator]

PS, can you upload the savelist and sensitivity control file you are using so that we can re-create your simulations exactly?

Tony.

[adjie]

thanks for your responses. i attach the process of making sensitivity test in

operasion in the sample file.docx

(87.63 KiB) Downloaded 209 times

. and, i get the negative result in iteration 1595. the negative value is not correct since i set the minimum value of 0.001in the formula. sorry, i can not attach the result as it is too big (17 MB).

thanks,

[tomfid]

I think the negative values you're finding are probably -1.29807e+033, which is the special value :NA:, which is filled in when simulation fails due to a floating point error.

The reason you're getting FP errors is that you've set an extremely narrow min/max range. The way the RANDOM functions work is to make a draw from the unconstrained distribution (which will be mean=1 in your case), then check whether the value lies within the range. If it doesn't, it discards the result and makes another draw. There's a limit to how many attempts it will make, after which it fails and reports a floating point error. So the process is likely to fail if the min/max range is narrow, especially if it's not near the mean, as in your case.

[tomfid]

From the Help for RANDOM functions:

NOTE If the minimum and maximum specified make it impossible for Vensim to find a number with the desired distribution (this occurs when more than 2000 attempts are discarded because of truncation) the function will return NA and an error message will be issued. The resulting return value will also likely cause a floating point error to occur in a downstream location which will stop the simulation. As a general rule of thumb the min and max should encompass 10% or more of the underlying probability distribution function to get reasonable results.

Narrow limits requiring many draws will also make the model slow.