mosaik.HDF5-SemVer 0.4.0rc20210914000526

Stores mosaik simulation data in an HDF5 database.

mosaik-hdf5

Store mosaik simulation data in an HDF5 database.

Mosaik-hdf5 stores the relations-graph of your simulation, timeseries for entities connected to it and optionally static entity and simulation meta data. The database structure usually looks like this:

/ [meta data]
|
+- Relations
|  |
|  +- Sim-0.Entity-1
|  |
|  +- PyPower-0.1-Node-2
|
+- Series
   |
   +- Sim-0.Entiy-1 [static data]
   |  |
   |  +- val_out
   |
   +- PyPower-0.1-Node-2 [static data]
      |
      +- P
      |
      +- Q

The Relations group contains one dataset for every entity. For each of the entity’s relations, the dataset has one tuple (path_to_relation, path_to_relatio_series).

The Series group contains (by default) one group for every entity. Each of these group as one dataset for every attribute.

Static entity data is stored as attributes in the entity groups. Simulation meta data is stored as attributes of the root group.

You can, optionally, create a more nested structure for the series, for example, if you want to group entities by simulator and/or simulator instance. This is done via regular expression replacements based on the entity ID.

Installation

mosaik-hdf5 uses the h5py module. If you get an error during installation that hdf5.h is missing, install the HDF5 headers (e.g., sudo apt-get install libhdf5-dev or brew install hdf5) or use a binary package (e.g., for Windows)

$ pip install mosaik.HDF5-SemVer

Usage

You can run mosaik-hdf5 as a sub-process or your simulation or in-process with it. Here are example configurations for both variants:

sim_config = {
    'HDF5-inproc': {
        'python': 'mosaik_hdf5:MosaikHdf5',
    },
    'HDF5-subproc': {
        'cmd': 'mosaik-hdf5 %(addr)s',
    },
}

Initialization

When you start mosaik-hdf5, you have to provide a step_size and a duration argument. The step_size defines how often data will be collected. The duration is the simulation end time in seconds. It is used to calculate the dataset size for every time series. For example, if duration is half-an-hour (1800s) and step_size is 60, each dataset will have a length of 30.

You can optionally pass a series_path tuple which contains a regular expression pattern and replacement string (see the Python docs for details).

For example, by default the entity IDs Sim-0.Entity-1 and PyPower-0.1-Node-2 would map to the series paths /Series/Sim-0.Entity-1 and /Series/PyPower-0.1-Node-2. But you want to group the entities by simulator type and simulator instance. Also, since one mosaik-pypower instance can contain multiple grids, you also want to take care of that. So what you want is something like that: /Series/Sim/Sim-0/Sim-0.Entity-1 and /Series/PyPower/PyPower-0.1/PyPower-0.1-Node-2. In this (rather complex) case, series_path can be (r'(((\w+)-(\d+.\d+|\d+))[.-](.*))', r'\3/\2/\1'). Easy, isn’t it?

Here are two examples for this:

a = world.start('HDF5', step_size=60, duration=1800)

pattern = r'(((\w+)-(\d+.\d+|\d+))[.-](.*))'
repl = r'\3/\2/\1'
b = world.start('HDF5', step_size=1, duration=10,
                series_path=(pattern, repl))

Model instantiation

Every instance of mosaik-hdf5 allows you to create exactly one instance of its Database model (which is also the only model provided). The Database has the following parameters:

• gridfile is the filename of the HDF5 database that will be created.

• buf_size (default: 1000) is the size of the internal data buffer for each series dataset. Mosaik-hdf5 buffers the data for every dataset and only writes larger chunks of data to the disk in order to improve the writing performance. If you have a lot of entities (> 100k) and only little memory, you may reduce this number. If you have lots of RAM, you can play with larger buffer sizes and see if it improves the performance for you.

• dataset_opts (default: None) is a dictionary of arguments that get passed to h5py’s create_dataset() method.

  This can, for example, be used to enable compression (note, that the lzf compression is not supported by all HDF5 viewers).

Examples:

# Basic usage
hdf5 = world.start('HDF5', step_size=1, duration=1)
db = hdf5.Database('data.hdf5')

# Use gzip compression
hdf5 = world.start('HDF5', step_size=1, duration=1)
db = hdf5.Database('data.hdf5', dataset_opts={
    'compression': 'gzip',
    'compression_opts': 9,
})

# Use lzf compression and a larger buffer
hdf5 = world.start('HDF5', step_size=1, duration=1)
db = hdf5.Database('data.hdf5', buf_size=1336,
                   dataset_opts={'compression': 'lzf'})

Storing data

The Database model has no attributes, but it accepts any inputs. This means that you can just connect anything to it. For each entity and attribute that is connected to the database, a corresponding dataset will be created in the database.

Mosaik-hdf5 also provides to extra methods that allow you to store some simulation meta data and static entity data. You can only use these methods once you created an instance of the Database model. The method set_meta_data() takes a single dict with an arbitrary amount of key-values pairs. The method set_static_data() takes a dict of entities and data dicts.

In the following example, we’ll create some (fake) PV entities and a power grid (with nodes and lines). We want to store the PV’s active and reactive power (P, Q), the node voltage and angle (Vm, Va) for all nodes and the complex current (I_real, I_imag) of all branches:

pv_pmax = 10
pvs = make_pvs(pv_pmax, ...)  # A list of PV entities
nodes, lines = make_grid(...)  # Lists of nodes/lines of a power grid

hdf5 = world.start('HDF5', step_size=1, duration=10)
db = hdf5.Database('data.hdf5')

# Store meta and static data
hdf5.set_meta_data({'duration': 10, 'description': 'hdf5 demo'}
hdf5.set_static_data({pv.full_id: {'p_max': pv_pmax} for pv in pvs})

# Connect inputs to database
mosaik.util.connect_many_to_one(world, pvs, db, 'P', 'Q')
mosaik.util.connect_many_to_one(world, nodes, db, 'Vm', 'Va')
mosaik.util.connect_many_to_one(world, lines, db, 'I_real', 'I_imag')

For a real example, you can take a look at the mosaik-demo.

Getting help

If you need, please visit the mosaik-users mailing list .

Changelog

0.3 - 2016-02-15

• [NEW] Implemented the new “setup_done()” method.

0.2 – 2014-10-29

• [NEW] More documentation

• [NEW] Static and simulation meta data can now be stored (issue #1).

• [NEW] Datasets can now be stored in arbitrarily defined paths (using complex regular expression replacements based on the entity ID) (issue #4).

0.1.2 – 2014-09-22

• [CHANGE] Updated to mosaik-api 2.0.

0.1.1 – 2014-07-31

• [FIX] Fixed a regression in 0.1.

0.1 – 2014-07-31

• Initial release

Authors

The mosaik HDF5 storage backend was created by Stefan Scherfke.