Statistical estimation of a clear sky signal from PV system power data
Project description
StatisticalClearSky
Statistical estimation of a clear sky signal from PV system power data
This project implements an algorithm based on Generalized Low Rank Models for estimating the output of a solar PV system under clear sky or "cloudless" conditions, given only measured power as an input. Noteably, no system configuration information, modeling parameters, or correlated environmental data are required. You can read more about this work in these two papers [1, 2].
We actually recommend that users generally not invoke this software directly. Instead, we recommend using the API provided by Solar Data Tools.
Getting Started
You can install pip package or Anaconda package for this project.
Recommended: Set up conda
environment with provided .yml
file
Updated September 2020
We recommend seting up a fresh Python virutal environment in which to use solardatatools
. We recommend using the Conda package management system, and creating an environment with the environment configuration file named pviuser.yml
, provided in the top level of this repository. This will install the solardatatools
package as well.
Please see the Conda documentation page, "Creating an environment from an environment.yml file" for more information.
Installation
If you are using pip:
$ pip install statisticalclearsky
As of February 11, 2019, it fails because scs package installed as a dependency of cxvpy expects numpy to be already installed. scs issue 85 says, it is fixed. However, it doesn't seem to be reflected in its pip package. Also, cvxpy doesn't work with numpy version less than 1.16. As a work around, install numpy separatly first and then install this package. i.e.
$ pip install 'numpy>=1.16'
$ pip install statisticalclearsky
If you are using Anaconda, the problem described above doesn't occur since numpy is already installed. And during statisticalclearsky installation, numpy is upgraded above 1.16:
$ conda install c slacgismo statisticalclearsky
Solvers
The default convex solver included with cvxpy is ECOS, which is open source. However this solver tends to fail on problems with >1000 variables, as it does not work for this algorithm.
So, the default behavior of the code is to use the commercial Mosek solver. Thus, we encourage you to install it separately as below and obtain the license on your own.

mosek  For using MOSEK solver.
If you are using pip:
$ pip install f https://download.mosek.com/stable/wheel/index.html Mosek
If you are using Anaconda:
$ conda install c mosek mosek==8.1.43
Academic licenses are available for free here: https://www.mosek.com/products/academiclicenses/
Usage
As a part of Python code or inside Jupyter notebook
Example 1: Simplest example with the fewest number of input parameters.
Using default solver (Open Source solver: ECOS)
import numpy as np from statistical_clear_sky.algorithm.iterative_fitting import IterativeFitting # Usually read from a CSV file or a database with more data, # covering 1 day (column) and a few years (row): power_signals_d = np.array([[0.0, 0.0, 0.0, 0.0], [1.33389997, 1.40310001, 0.67150003, 0.77249998], [1.42349994, 1.51800001, 1.43809998, 1.20449996], [1.52020001, 1.45150006, 1.84809995, 0.99949998]]) iterative_fitting = IterativeFitting(power_signals_d) iterative_fitting.execute() clear_sky_signals = iterative_fitting.clear_sky_signals() degradation_rate = iterative_fitting.degradation_rate()
Example 2: Estimating clear sky signals without degradation.
You can estimate clear sky signals based on the assumption that there is no yeartoyear degradation. In this case, you can set is_degradation_calculated keyword argument to False in execute method. By default, it's set to True.
import numpy as np from statistical_clear_sky.algorithm.iterative_fitting import IterativeFitting # Usually read from a CSV file or a database with more data, # covering 1 day (column) and a few years (row): power_signals_d = np.array([[0.0, 0.0, 0.0, 0.0], [1.33389997, 1.40310001, 0.67150003, 0.77249998], [1.42349994, 1.51800001, 1.43809998, 1.20449996], [1.52020001, 1.45150006, 1.84809995, 0.99949998]]) iterative_fitting = IterativeFitting(power_signals_d) iterative_fitting.execute(is_degradation_calculated=False) clear_sky_signals = iterative_fitting.clear_sky_signals()
Example 3: Using a different solver.
The default solver ECOS is not stable with large set of input data. The following example shows how to specify to use Mosek solver by passing solver_type keyword argument (to the constructor).
import numpy as np from statistical_clear_sky.algorithm.iterative_fitting import IterativeFitting # Usually read from a CSV file or a database with more data, # covering 1 day (column) and a few years (row): power_signals_d = np.array([[0.0, 0.0, 0.0, 0.0], [1.33389997, 1.40310001, 0.67150003, 0.77249998], [1.42349994, 1.51800001, 1.43809998, 1.20449996], [1.52020001, 1.45150006, 1.84809995, 0.99949998]]) iterative_fitting = IterativeFitting(power_signals_d, solver_type='MOSEK') iterative_fitting.execute() clear_sky_signals = iterative_fitting.clear_sky_signals() degradation_rate = iterative_fitting.degradation_rate()
Example 4: Setting rank for Generalized Low Rank Modeling.
By default, rank of low rank matrices is specified to be 6. You can change it by specifying rank_k keyword argument (in the constructor).
import numpy as np from statistical_clear_sky.algorithm.iterative_fitting import IterativeFitting # Usually read from a CSV file or a database with more data, # covering 1 day (column) and a few years (row): power_signals_d = np.array([[0.0, 0.0, 0.0, 0.0], [1.33389997, 1.40310001, 0.67150003, 0.77249998], [1.42349994, 1.51800001, 1.43809998, 1.20449996], [1.52020001, 1.45150006, 1.84809995, 0.99949998]]) iterative_fitting = IterativeFitting(power_signals_d, rank_k=6) iterative_fitting.execute() # Get the resulting left low rank matrix and right low rank matrix for evaluation. left_low_rank_matrix = iterative_fitting.left_low_rank_matrix() # The above can be also obtained as l_cs_value: l_cs_value = iterative_fitting.l_cs_value # Get the resulting right low rank matrix for evaluation. right_low_rank_matrix = iterative_fitting.right_low_rank_matrix() # The above can be also obtained as r_cs_value: r_cs_value = iterative_fitting.r_cs_value clear_sky_signals = iterative_fitting.clear_sky_signals() degradation_rate = iterative_fitting.degradation_rate() # The above can be also obtained as beta_value: beta_value = iterative_fitting.beta_value
Example 5: Setting different hyperparameters for minimization of objective function of Generalized Low Rank Modeling.
There are three hyperparameters in the objective function of Generalized Low Rank Modeling, i.e. mu_l, mu_r, and tau. By default, mu_l is set to 1.0, mu_r is set to 20.0, and tau is set to 0.8. You can change it by specifying mu_l, mu_r, and tau keyword arguments in execute method.
import numpy as np from statistical_clear_sky.algorithm.iterative_fitting import IterativeFitting # Usually read from a CSV file or a database with more data, # covering 1 day (column) and a few years (row): power_signals_d = np.array([[0.0, 0.0, 0.0, 0.0], [1.33389997, 1.40310001, 0.67150003, 0.77249998], [1.42349994, 1.51800001, 1.43809998, 1.20449996], [1.52020001, 1.45150006, 1.84809995, 0.99949998]]) iterative_fitting = IterativeFitting(power_signals_d) iterative_fitting.execute(mu_l=5e2, mu_r=1e3, tau=0.9) clear_sky_signals = iterative_fitting.clear_sky_signals() degradation_rate = iterative_fitting.degradation_rate()
Example 6: Setting different control parameters for minimization of objective function of Generalized Low Rank Modeling.
There are three control parameters in the objective function of Generalized Low Rank Modeling, i.e. exit criteria  exit_criterion_epsilon, and maximum number of iteration  max_iteration. By default, exit_criterion_epsilon is set to 1e3, max_iteration is set to 100. You can change it by specifying eps and max_iteration keyword arguments in execute method.
import numpy as np from statistical_clear_sky.algorithm.iterative_fitting import IterativeFitting # Usually read from a CSV file or a database with more data, # covering 1 day (column) and a few years (row): power_signals_d = np.array([[0.0, 0.0, 0.0, 0.0], [1.33389997, 1.40310001, 0.67150003, 0.77249998], [1.42349994, 1.51800001, 1.43809998, 1.20449996], [1.52020001, 1.45150006, 1.84809995, 0.99949998]]) iterative_fitting = IterativeFitting(power_signals_d) iterative_fitting.execute(exit_criterion_epsilon=1e6, max_iteration=10) clear_sky_signals = iterative_fitting.clear_sky_signals() degradation_rate = iterative_fitting.degradation_rate()
Example 7: Setting limit on degradation rate.
You can specify the maximum degradation and minimum degradation by setting max_degradation and min_degradation keyword arguments in execute method. By default, they are set not to be used.
import numpy as np from statistical_clear_sky.algorithm.iterative_fitting import IterativeFitting # Usually read from a CSV file or a database with more data, # covering 1 day (column) and a few years (row): power_signals_d = np.array([[0.0, 0.0, 0.0, 0.0], [1.33389997, 1.40310001, 0.67150003, 0.77249998], [1.42349994, 1.51800001, 1.43809998, 1.20449996], [1.52020001, 1.45150006, 1.84809995, 0.99949998]]) iterative_fitting = IterativeFitting(power_signals_d) iterative_fitting.execute(max_degradation=0.0, min_degradation=0.5) clear_sky_signals = iterative_fitting.clear_sky_signals() degradation_rate = iterative_fitting.degradation_rate()
Jupyter notebook examples
Alternatively, you can clone this repository (GIT) and execute the example codes under notebooks folder.
Simplest way to install dependencies if you are using pip is by
$ pip install r requirements.txt
As mentioned in the section, "Getting Started" above, as of February 11, 2019, it fails because scs package installed as a dependency of cxvpy expects numpy to be already installed. scs issue 85 says, it is fixed. However, it doesn't seem to be reflected in its pip package. Also, cvxpy doesn't work with numpy version less than 1.16. As a work around, install numpy separatly first and install the other packages using requirements.txt. i.e.
$ pip install 'numpy>=1.16'
$ pip install r requirements.txt
Running the tests
Unit tests (developer tests)

GIT clone this project.

In the project directory in terminal,
$ python m unittest
This runs all the tests under tests folder.
All the tests are placed under "tests" directory directly under the project directory. It is using "unittest" that is a part of Python Standard Library by default. There may be a better unit testing framework. But the reason is to invite as many contributors as possible with variety of background.
Coding style tests
pylint is used to check if coding style is conforming to "PEP 8  Style Guide for Python Code"
Note: We are open to use LGTM. However, since we decided to use another code coverage tool codecov based on a comment by project's Technical Advisory Council, we decided not to use another tool that does code coverage. We are also open to use other coding style tools.
Example of using pylint:
In the project directory in terminal,
$ pylint statistical_clear_sky
Contributing
Please read CONTRIBUTING.md for details on our code of conduct, and the process for submitting pull requests to us.
Versioning
We use Semantic Versioning for versioning. For the versions available, see the tags on this repository.
Authors

Bennet Meyers  Initial work and Main research work  Bennet Meyers GitHub

Tadatoshi Takahashi  Refactoring and Packaging work and Research support work  Tadatoshi Takahashi GitHub
See also the list of contributors who participated in this project.
License
This project is licensed under the BSD 2Clause License  see the LICENSE file for details
References
[1] B. Meyers, M. Tabone, and E. C. Kara, "Statistical Clear Sky Fitting Algorithm," IEEE Photovoltaic Specialists Conference, 2018.
Acknowledgments
 The authors would like to thank Professor Stephen Boyd from Stanford University for his input and guidance and Chris Deline, Mike Deceglie, and Dirk Jordan from NREL for collaboration.
Project details
Release history Release notifications  RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Filename, size  File type  Python version  Upload date  Hashes 

Filename, size statistical_clear_sky0.4.4py3noneany.whl (45.8 kB)  File type Wheel  Python version py3  Upload date  Hashes View 
Filename, size statisticalclearsky0.4.4.tar.gz (36.9 kB)  File type Source  Python version None  Upload date  Hashes View 
Hashes for statistical_clear_sky0.4.4py3noneany.whl
Algorithm  Hash digest  

SHA256  8dfe81e30f013eafa81c79395119f5ef6ba793fb0d1849da601add68be14dcdc 

MD5  181ae03f5ef711ddbf100f3881f09662 

BLAKE2256  7f10e859afcd7c82f9e9f2752122239424ba2ba8fcabb3e97166404650cf8f7b 
Hashes for statisticalclearsky0.4.4.tar.gz
Algorithm  Hash digest  

SHA256  3a35c8247c05a7841fdc5e233bd86d8be26294d82f9240a6f67cd42d823b1b52 

MD5  9302df39ea3ced1cd705903feb32c59e 

BLAKE2256  78b99bf31f28ca4fdcfda01bd2cbed2a171f4855718732a5e734826dda9fcfdf 