The first part of this page provides an installation guide, see Installation. Second, an overview of the architecture is presented and the core features and components are discussed, see Architecture. At last, it is exemplary illustrated how windML is used, see Running an Example.
Before the installation of windML you have to make sure that all needed Dependencies are installed. In order to install windML, you have to check out a working copy of our stable branch in our development repository.
$ git clone https://github.com/cigroup-ol/windml.git
After cloning the stable branch, the new folder windml is located in your current working directory. Make sure your windML-folder is in your Python PATH by executing.
$ export PYTHONPATH=$PYTHONPATH:<windml-directory>
Below, an illustration of the windML architecture is presented. At the top, one can see exemplary available data sources of wind data time-series open to the public on the internet. The DataSource classes implemented in windML download the data from data mirrors, parse the data into an windML-specific data format and keep the data in a local cache. The windML-specific format is defined by the Windpark and Turbine classes. See Wind Park and Turbine documentation for the windML-specific model. The wind parks and turbines are selected by specifying ID and radius. See Datasets page for the documentation of the data sets and methods to fetch time-series.
Given Windpark and Turbine objects, one can visualize the data via different visualization components such as dimensionality reduction, park and turbine information, information about the time-series, topology etc., see examples. An important motivation of windML is forecasting time-series with regression and classification. For both methodologies, a mapping of a time-series to labels is required. Different mapping approaches have been tested in the past, see Mapping for explanation of the various mapping methods. In the current release, the following regression techniques have been applied: support vector regression (SVR), KNN regression, and linear regression.
Running some examples of windML is probably the best way to start. In the examples gallery all scripts from the /examples folder of the windML installation are plotted. In order to run an example, one only has to execute the corresponding Python script. Please make sure to install windML correctly, see the Installation page.