EnCortex-Framework
Energy Arbitrage
In this scenario, a Renewable Energy provider has a Lithium-ion Batteries to it’s customers. Now, these batteries are operated(i.e. charged, discharged or maintained-idle) throughout it’s operational period in order to maximize cost savings(increase cost savings by charging/discharging the battery at off-peak/peak price periods), carbon emissions savings(by charging/discharging when usage of renewable energy as energy source is high/low) or a weighted-average of both.
Thus, it is important to optimize the management and scheduling of these batteries (when to charge or discharge) to maximize the producer’s objective.
Brief look into EnCortex implementation details
- The Entities used here are
BatteryandGrid, which is an umbrella term for a utility grid. - The action space or the decision in this scenario is to either charge the battery, discharge the battery or letting it stay idle.
- The supported class of optimizers are
Mixed-Integer Linear ProgrammingandReinforcement Learning - The data in this scenario is only present with the
Gridentity. Thecsvfile format is illustrated in the table below:
:header-rows: 1
* - timestamps
- prices
- emissions
* - 2019-10-20 00:00:00
- 40.47
- 116.0
* - ...
- ...
- ...
For more details, please refer to Section 5.1 in our paper linked below:
```{button-link} ../_static/nsdi23fall-paper705.pdf
:color: primary
:outline:
EnCortex paper
Here, we aim to use the below objective function:
````{eval-rst}
.. math::
:label: Objective function for Energy Arbitrage
max \sum_{t=1}^{T}(\omega_{carbon}{Em}_{t} + \omega_{cost}{Cost}_{t} - \omega_{deg}{DegCost}_{t})
where $\omega_{field}$ corresponds to relative weight/importance of that field/attribute. The field here are cost savings($Cost_t$), carbon emission savings($Em_t$) and degradation cost associated with the battery($DegCost_t$). The $T$ corresponds to the horizon over which the objective function is maximized.
Running Energy Arbitrage
To run the Energy Arbitrage experiment, you’ll need a specific config.yaml file. Create the config.yml in your root folder of your AML workspace(as shown below).
battery_arbitrage_config.yml(click to open/close)
```yaml #Li-Ion Battery Parameters storage_capacity: 10. efficiency: 1. soc_initial: 1. depth_of_discharge: 90. soc_minimum: 0.1 time_unit: 15 milp_flag: true solver: "ort" test_flag: false min_battery_capacity_factor: 0.8 battery_cost_per_kWh: 20. reduction_coefficient: 0.99998 degradation_period: 7 weight_emission: 1.0 weight_degradation: 0.0 weight_price: 1.0 degradation_flag: false seed: 40 ```Explanation of the above configuration values(click to open)
```{list-table} :header-rows: 1 * - Parameter - Description - Value * - storage_capacity - Battery Storage capacity(kWh) - 10. * - efficiency - Battery Charge/Discharge efiiciency - 1. * - soc_initial - SoC initial(Between 0-1) - 1. * - depth_of_discharge(%) - Depth of Discharge - 90. * - soc_minimum(between 0-1) - SoC minimum - 0.1 * - time_unit - Minimum time unit in mins - 15 * - milp_flag - Use MILP - true * - solver - Solver to use['ort','dqn'(RL)] - "ort" * - test_flag - False - false * - min_battery_capacity_factor - Minimum Battery Capacity Factor - 0.8 * - battery_cost_per_kWh - Battery Cost per kWh - 20. * - reduction_coefficient - Battery reduction coefficient - 0.99998 * - degradation_period - Degradation period in days - 7 * - weight_emission - Emission saving weightage(between 0-1) - 1.0 * - weight_degradation - Degradation weightage(between 0-1) - 0.0 * - weight_price - Cost saving weightage(between 0-1) - 1.0 * - degradation_flag - Flag to enable battery degradation - false * - seed - Seed of the experiment - 40 ```Running via CLI
To run locally, run the following command on your EnCortex docker container bash(if using Docker container) or your terminal/jupyter notebook cell(if installed via pip)
python examples/battery_arbitrage_example.py
Visualizations through the CLI can only be viewed when run locally(through Streamlit). The final numbers are visible in the logs.
Other variations of the configuration can be found below. In the tab mentioned Degradation(enabled), the degradation of a battery during charge/discharge cycles is considered and is enabled by the setting flag degradation_flag as true.
.. tabs::
.. tab:: Degradation(disabled)
**config.yml**
.. code:: yaml
storage_capacity: 10.
efficiency: 1.
soc_initial: 1.
depth_of_discharge: 90.
soc_minimum: 0.1
time_unit: 15
milp_flag: true
solver: "ort"
test_flag: false
min_battery_capacity_factor: 0.8
battery_cost_per_kWh: 20.
reduction_coefficient: 0.99998
degradation_period: 7
weight_emission: 1.0
weight_degradation: 0.0
weight_price: 1.0
degradation_flag: true
seed: 40
.. tab:: Degradation(enabled)
**config.yml**
.. code:: yaml
storage_capacity: 10.
efficiency: 1.
soc_initial: 1.
depth_of_discharge: 90.
soc_minimum: 0.1
time_unit: 15
milp_flag: true
solver: "ort"
test_flag: false
min_battery_capacity_factor: 0.8
battery_cost_per_kWh: 20.
reduction_coefficient: 0.99998
degradation_period: 7
weight_emission: 1.0
weight_degradation: 0.0
weight_price: 1.0
degradation_flag: false
seed: 40
Running on an AML notebook
We also provide a notebook to ease modification and help visualize your results(click on the button below).
```{button-link} ../notebooks/energy_arbitrage.ipynb :color: primary :shadow:
Energy Arbitrage Notebook
To run the notebook:
1. Download the notbook from this documentation(download button {octicon}`download;1em;sd-text-info` on the top of the page) and upload the notebook to AML studio
2. Connect to your compute instance(see [setup](setup/azureml/compute))
3. Select `Python3.8 - AzureML` as your environment
4. Run the pip instructions to install `EnCortex` (see [setup](setup/pip))
> Results and visualizations can be viewed directly on the notebook.
````{margin}
```{note}
This is a one-time setup. As long as the compute environment is not changed, this step can be ignored after the first installation.
````