RIPE: Reaction Identification and Parameter Estimation¶
The RIPE module provides tools for reaction network identification. RIPE uses reactor data consisting of concentration, or conversion, values for multiple species that are obtained dynamically, or at multiple process conditions (temperatures, flow rates, working volumes) to identify probable reaction kinetics. The RIPE module also contains tools to facilitate adaptive experimental design. The experimental design tools in RIPE require the use of the python package RBFopt. More information for RBFopt is availible at www.github.com/coin-or/rbfopt
Basic Usage¶
RIPE can be used to build models for static datasets through the function ripe.ripemodel
ripe_results = ripe.ripemodel(data, kwargs)
data is provided to RIPE as one, two, or three dimensional python data structures, where the first axis corresponds to observations at different process conditions, the second axis corresponds to observations of different chemical species, and the third axis corresponds to dynamic observation of a chemical species at a specified process condition.
RIPE adaptive experimental design can be accessed using ripe.ems
[proposed_x, errors] = ripe.ems(ripe_results, simulator, l_bounds, u_bounds, n_species, kwargs)
ripe_results - The results from ripe.ripemodel, additional information provided in the results section
simulator - a black-box simulator for the unknown process.
l_bounds/u_bounds - lower and upper bounds for the input variables in the adaptive design
nspecies - the number of chemical species present in the black-box system
Reaction stoichiometries and mechanisms are provided explicitly to ripemodel through the keyword arguments mechanisms and stoichiometry. Detailed explanations of the forms of these arguments are provided in the stoiciometry and mechanism specification section. Additional keyword arguments can be found in the additional options section.
RIPE Output¶
By default, one file will be generated
riperesults.txt - a file containing the selected reactions and parameter estimates
Reaction Stiochiometry and Mechanism Specification¶
Considered reaction stiochiometries are provided through keyword arguments.
Stoichiometry
Considered reaction stoichiometries are defiend as a list of list, where reactants and products are defined as negative and positive integers , respectively, according to their stoichiometric coefficeints. A set of considered reaction stoichiometries must be provided. If process data consists of species conversion, a positive coefficient should be specified.
Mechanisms
Considered reaction mechanisms are provided explicitly to RIPE through q keyword argument. If no kinetic mechanisms are specified, mass action kinetics are ascribed to every considered stoichiometry. RIPE contains kinetic mechanisms defined internally, and called through ripe.mechs.<mechanism>. The availible mechanisms include:
massact - mass action kinetics, order informed by reaction stoichiometry
19 empirical rate forms included relate specifically to catalyst conversion in chemical looping combustion reactors include:
Random nucleation
Power law models
Avrami-Erofeev models
These internal kinetics can be specified by calling ripe.mechs.massact or ripe.mechs.clcforms respectively. User-defined kinetic mechanisms can also be supplied to RIPE as python functions. An example is provided in the file crac.py.
Additional Results and Options¶
In addition to the arguments stoichiometry and mechanism, a number of other optional arguments are availible, including:
Arguments relating to process conditions
x0 - initial concentration at each process condition for every species
time - time associated with dynamic samples for every process condition
temp - temperature associated with every process condition
flow - flow rate at every process condition for every species
vol - reactor volume at every process condition
Arguments related to RIPE algorithmic function
tref - reference termpeature for reformulated Arrhenius models
ccon - specified cardinality constraint instead of BIC objective
sigma - expected variance of noise, estimated if not provided
onemechper - one mechanism per stoichiometry in selected model, true by default
Additional arguments
minlp_path - path to baron or other minlp solver, can also be set in shared.py
alamo_path - path to alamo, can also be set in shared.py
expand_output - provide estimates for noise variance in model resutls
zscale - linear scaling of observed responses between -1 and 1
ascale - linear scaling of activities between -1 and 1
hide_output - surpress output to terminal
keepfiles - keep scratch files for debugging
showpyomo - show pyomo output to terminal, false by default
RIPE Examples¶
Three examples are included with RIPE. These examples demonstrate different use cases, and provide a template for utilizing user-defined mechanisms.
clc.py - a chemical looping combustion example in which catalyst conversion is observed over time
isoT.py - an example that utilizes both ripe.ripemodel and ripe.ems
crac.py - an example that utilizes user-defined reaction mechanisms
All of these examples are built for Linux machines. They can be called from the command line by calling python directly, or can be called from inside a python environment using execfile().