What does changing seed value in Algorithm Settings do?

asked 2019-12-05 20:34:59 -0600

y.tanaka
213 ●1 ●6

updated 2020-01-07 17:33:36 -0600

__AmirRoth__

4446 ●5 ●17 http://bleedinggreenna...

Context:

I am working with project_singlefamilydetached from the ResStock project
I have location.tsv and location-region.tsv fixed to only simulate homes in Alpena, Michigan
I have changed the vintage.tsv for Alpena, Michigan to my own probabilities of my choosing

I have been using PAT tool to simulate differences in energy use for homes with a Propane Furnace only vs. GSHP only. However, I noticed that for different seeds, the building characteristics remain identical after analysis has finished.

An example is shown below: image description

The average, annual total heating costs for 5 different analyses, with 5 unique seeds, produce almost identical values. To produce the results you see before you, I utilized the electricity_heating_kwh and propane_heating_mbtu outputs from the enduse_timeseries.csv for each of the buildings.

If changing the seeds does not change the outputs of enduse_timeseries.csv, then, what does it change? I wish to create a distribution of comparisons, but in order to do that, the results outputted by PAT needs to change as well.

edit retag flag offensive close merge delete

add a comment

answered 2019-12-05 23:51:17 -0600

shorowit

11931 ●203 ●9 https://www.nrel.gov/b...

ResStock projects do not use this PAT interface input.

Note that ResStock's sampling algorithm is part deterministic and part random. The deterministic part occurs when ResStock determines how many of a given option to choose for a given building characteristic. For example, if a probability distribution for wall insulation is 30% uninsulated, 50% R-13, and 20% R-19, and you asked for 20 buildings, ResStock's sampling would always yield 6 buildings with uninsulated walls, 10 buildings with R-13 walls, and 4 buildings with R-19 walls.

Randomness thus only comes into play when these wall options are combined with other parameters (HVAC options, appliance options, etc). However, ResStock currently uses a hard-coded random seed to make the sampling algorithm repeatable. If you really want to generate different combinations of building characteristics, you can change this value between runs.

Comments