Page 1 of 1

Monte Carlo & parameter variation of product streams

Posted: 20 Jun 2016 11:17
by bbuchspies

I am modelling bioenergy systems in which outputs are dependent on uncertain input variables (e.g. biomass compositon).

I am currently struggeling to account for uncertainties related to product streams or parameterized product streams:
Let's say the model of interest consists of a single process with 2 products (A&B) and an elementary flow entering the system (E_in, e.g. resources) and an elementary flow leaving the system (E_out, e.g. emission).
The monte carlo analysis accounts for uncertainties that are defined for E_in and E_out.
However, as far as I found out, openLCA is not able to account for uncertainties related to each product (A & B), as calculated allocation factors remain fixed.
To do so, allocation factors need to be calculated for each run of the monte carlo simulation to account for differing quantities of products.

Another option, parameter variation, seems not to help either:
A paramter variation could help to test a couple of different variants. Using parameter variation (in a project) does not help either as allocation factors remain fixed...
So the only way as see at the moment is to copy all processes affected and to change parameters manually first and update allocation factors afterwards.
This very time consuming and means that the parameter variation and uncertainty analysis, as implemented in openLCA, is not suitable to account for uncertain or parameterized product streams.

Is there any way to solve this problem?

Re: Monte Carlo & parameter variation of product streams

Posted: 22 Jun 2016 08:51
by aciroth
Hi, interesting question but I am not sure whether I fully understood: for allocation, always "the other" product is not considered in the product system and calculation.Therefore you will not see the two products A and B in your result but only e.g. A. If you parameterise the amount for the product you are interested in, and e.g. add uncertainty to the parameter, then this will be reflected in the MC simulation. I am not sure whether this answers your question?
Best wishes,

Re: Monte Carlo & parameter variation of product streams

Posted: 24 Jun 2016 16:27
by bbuchspies

thank you for your fast response, Andreas. I was refering to another aspect:
If the outputs of my process are dependent of input paramters (e.g. biomass composition), the quantity of output changes according to the input paramters. If all output streams change linearly with varying input paramters, allocation factors are not affected. However, if they change in another manner, allocation factors for each product need to be assessed for each variant of input paramters.

When using the 'project' function in openLCA, allocation parameters remain fixed, as defined in the process of interest based on the parameter setting in the process itself.

We consider a process with two products "fuel" (reference product) & "feed" (co-product); input parameters "A" & "B", CO2 emissions:

In: | Out:
(inputs) | fuel= A
(inputs) | feed= A*B
(inputs) | Carbon dioxide, fossil

Now, we define A=2 and B=3. This results in (physical) allocation factors of 0.25 (fuel) and 0.75 (feed).
If I apply parameter variation in a 'project', i.e. test a variation of A & B (e.g. A=3, B=5), allocation factors for "feed" and "fuel" need to be recalculated by openLCA.
However, this is not done. openLCA keeps the allocation factors 0.25 and 0.75 based on A=2 and B=3, as defined in the process.

The same applies to product streams for which an uncertainty distribution is defined. For each run of the monte carlo simulation, the quantity of outputs changes and, thus, the allocation factor for each product, as the ratio of output changes in a non-linear way.
So allocation factors need to be re-calculated for each run of the simulation. This is not done by openLCA either:

Considering the previous example, there are four possibilites of defing or not-defining uncertainties for products "fuel" and "feed".
a) Uncertainty disribution for both (e.g. normal distribution).
b) Unertainty distribution for fuel (=reference product), no uncertainty distribution defined for feed.
c) Uncertainty distribution for feed, no uncertainty distribution defined for fuel.
d) no uncertainties defined. (not of interest here)

Monte Carlo Analysis of these cases gives a distribution according the respective distribution that was defined (e.g. normal distribution) in case a) and b).
No distribution is calculated for case c). MC analysis results in a single bar!

That indicates that the resuting distribution of a MC analysis originates solely from the ratio 'emissions-to-reference product. (e.g. kg CO2 per MJ of fuel) and is not influenced by changing allocation factors resulting from the ratio reference product-to-co-product.

I hope that these examples help to understand the previsously posted problem.

Thanks for your help!