frank1010111 / pywaterflood Goto Github PK

Add vEH water influx model for predicting water production from aquifers. There are linear and radial flow models.

1. The "simple example" is also a bit underwhelming (although the notebook is much better). It produces no graphics or table to visualize the results, print anything to the screen, or provide any feedback to users.
2. The writing is a bit terse, even for JOSS. Please see the general remarks below and the comments in the attached paper.
3. Please add a note to the main readme explaining how to run the tests. I recall this is strongly suggested by JOSS at the submission stage, and seems it should be included for anyone who wishes to contribute to the software or develop on top of it.
4. I suggest not describing a method as having been "perfected".

Shortened from review originally posted by @mgcooper in openjournals/joss-reviews#6191 (comment)

Is your feature request related to a problem? Please describe.
It would be nice to show how you can use CRM results to optimize water injection

Describe the solution you'd like

1. Start out with production and injection histories
2. Run CRM to get the gains from injector to producer
3. Set a water injection budget, limits on how high or low each injector can go, etc
4. Optimize oil production given these constraints

Describe alternatives you've considered
This could become a feature of the library if it gets enough interest.

Additional context
See https://www.sciencedirect.com/science/article/pii/S0920410509002046 and other examples from literature

Is your feature request related to a problem? Please describe.
I like to see the connectivity between injectors and producers at a map level.

Describe the solution you'd like
Given a fit CRM model and injector and producer locations, there should be a matplotlib quiver plot. with arrows given lengths and colors commensurate with the connectivity. This should be in a new submodule importable through pywaterflood.plotting.

Additional context
A demo plot looks like this

Is your feature request related to a problem? Please describe.
I would like to use this package in python 3.12

Describe the solution you'd like
New wheels and testing for python 3.12

Describe alternatives you've considered
I could use python 3.11, but why not the newest?

Additional context
With python 3.12 released, we're one release closer to python-pi! 3.14 is only two years away.

Add python 3.11 compatibility to pyproject.toml and test it with github actions. Simplify any dangling dependencies.

Is your feature request related to a problem? Please describe.
I have tested the library, and I noticed that $$\sum^{N_{prod}}_{j=1} fij\leq1$$ for a fixed injector $i$ is not fulfilled. This is the constrain given by Eq. 19 in Holanda et al. 2018. I have tried every option for constraints in CRM(constraints=..) and the issue remains. Not sure if I have missed/did something wrong and the constraint is already implemented.

Describe the solution you'd like
It would very helpful to have such constraint implemented. Make the solution more realistic, since in the real oil field, injection losses occur.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

Is your feature request related to a problem? Please describe.
While the CRM can forecast production, there are no notebooks with examples of this

This would be added to the docs/user-guide/ as another notebook.

Describe the bug
Readthedocs is dying during builds due to a "AttributeError: 'Module' object has no attribute 'doc'" presumably related to multiwellproductivity.

To Reproduce
Steps to reproduce the behavior:
This is part of the CI/CD pipeline. It happens on each change to master.

Expected behavior
It passed until this pull request: https://github.com/frank1010111/pywaterflood/pull/57/files. Nox runs still pass.

nox -s docs

builds on Linux

I have successfully installed pywaterflood and can run 7-minutes-to-pywaterflood.ipynb, choosing-crm.ipynb, and multiwell-productivity-index.ipynb. However, I can't run /buckley-leverett.ipynb

This is the error I get when trying to run the first cell in the notebook.

ModuleNotFoundError Traceback (most recent call last)
Cell In[3], line 6
3 import seaborn as sns
4 import pandas as pd
----> 6 from pywaterflood.buckleyleverett import Reservoir, water_front_velocity, breakthrough_sw

ModuleNotFoundError: No module named 'pywaterflood.buckleyleverett'

There are tutorials for CRM with and without pressure compensation, but not MPI. Let's fix that!

Is your feature request related to a problem? Please describe.
There's information in how the water/oil ratio is changing. We ought to capture that with a Koval fractional flow model. We can compare it to the Buckley-Leverett solutions.

Describe the solution you'd like
The equation

$$f_w=\begin{cases} 0, & t_D < 1/K_v \\ \frac{K_v - \sqrt{K_v/t_D}}{K_v - 1}, & 1/K_v \le t_D \le K_v \\ 1, & t_D > K_v \end{cases}$$

where $K_v = H_k E$, $H_k$ is the heterogeneity, $E = \left(0.78 + 0.22 \left(\mu_o/\mu_w\right)^{1/4}\right)^4$ is the effective viscosity ratio.

Describe alternatives you've considered
Buckley Leverett with straight-line relative permeabilities might be able to do this.

Further reading: https://doi.org/10.2118/217973-PA

Is your feature request related to a problem? Please describe.
It would be kind of fun to have an option to penalize gains/connectivities between injectors and producers

Describe the solution you'd like
An option when creating a CRM to use a penalty function, either L1 or L2, on the gains when fitting.

Additional context
Elastic net regression

Add solution for $S_w$ frontal advance

$$\begin{equation} \left(\frac{dx}{dt}\right)_{S_w} = \frac{q_t}{\phi A} \left(\frac{\partial f_w}{\partial S_w}\right)_t \end{equation}$$

where

$$\begin{equation} f_w = \frac{1 + \frac{k_o A}{\mu_o q_t} \left(\rho_o - \rho_w\right) g \sin \alpha}{1 + \frac{k_o}{k_w}\frac{\mu_w}{\mu_o}} \end{equation}$$

and $k_o$ and $k_w$ are dependent upon water saturation following the Brooks and Corey model.

$$\begin{equation} k_{ro} = \left(\frac{S_o- S_{or}}{1 - S_{or} - S_{wc}- S_{gc}}\right)^{n_o} \end{equation}$$

This project could use more and better documentation. It could also use fancier docs like this: https://github.com/frank1010111/petrelpy/tree/master/docs

Tests are failing due to numpy array size changes.

See https://github.com/frank1010111/pywaterflood/runs/6540148275?check_suite_focus=true

The message is

  E   ValueError: numpy.ndarray size changed, may indicate binary incompatibility. Expected 96 from C header, got 88 from PyObject

The Journal of Open Source Software ought to love this. Well, maybe.

Is your feature request related to a problem? Please describe.
When producers are shut in and not producing, their production is still predicted. Those lengths of time should not factor into the residuals for finding connectivity between the shut in producers and the active injectors.

Describe the solution you'd like
There should be a flag that, when set, changes the behavior so that when production is zero, the residual for a fit should be set to zero.

class CRM:
    ...
    def __init__(
        self,
        primary: bool = True,
        tau_selection: str = "per-pair",
        constraints: str = "positive",
        producer_shutins: bool = False,
    ):

Describe alternatives you've considered
A mask could be applied. This effect can be ignored (but maybe shouldn't).

Additional context
The function of interest is at

Bottomhole pressure variation at each producer can also affect its production rate. Yousef et al (2006) have a formulation of that contribution to production, which is

Kaviani et al (2012) suggest that a shift to production can be performed in segmented capacitance modeling with the equation

They then suggest it can be further simplified through assuming very large τ values between producers, which gets to the equation

The fitting parameters here are elements of v. This could either be solved simultaneously with the primary production rates and producer-injector connectivities, or as a pre-processing step.