This library is a small layer above brightway2, designed for the definition of parametric inventories with fast computation of LCA impacts, suitable for monte-carlo analyis.
lca-algebraic provides a set of helper functions for :
If you already have Anaconda & Jupyter installed, you can install the library with either pip or conda :
conda install -c oie-minesparistech lca_algebraic
pip install lca_algebraic
Alternatively, you can download and execute this installer. It will setup a full anaconda environment with Jupyter, Brightway2 and LCA Algebraic.
The full API is documented here.
It is distributed under the BSD licence.
The main idea of this libray is to move from procedural definition of models (slow and prone to errors) to a declarative / purely functionnal definition of parametric models (models as pure functions).
This enables fast computation of LCA impacts. We leverage the power of symbolic calculus provided by the great libary SymPy.
We define our model in a separate DB, as a nested combination of :
The amounts in exchanges are expressed either as static amounts, or symbolic expressions of pre-defined parameters.
Each activity of our root model is defined as a parametrized combination of the foreground activities, which can themselves be expressed by the background activities.
When computing LCA for foreground models, the library develops the model as a combination of only background activities. It computes once for all the impact of background activities and compiles a fast numpy (vectorial) function for each impact, replacing each background activity by the static value of the corresponding impact.
By providing large vectors of parameter values to those numpy functions, we can compute LCA for thousands of values at a time.
Under the hood, the activities we define with lca-algebraic are standard brightway2 activities. The amounts of exchanges are stored as float values or serialized as string in the property formula.
Parameters are also stored in the brightay2 projets, making it fully compatible with brightway.
Thus, a model defined with lca-algebraic is stored as a regular bw2 projet. We can use bw2 native support for parametrized dataset for computing LCAs, even if much more slower than the method explain here.