PIDE: Photovoltaic Integration Dynamics and Efficiency for Autonomous Control on Power Distribution Grids

This repository contains the Python implementation and resources to reproduce our work titled: PIDE: Photovoltaic Integration Dynamics and Efficiency for Autonomous Control on Power Distribution Grids. (The paper is currently under review in Energy Informatics journal as part of the collection "Digital Innovations for Sustainable and Resilient Energy Systems".)

---

1. Repository Overview

The provided Python-based framework, PIDE, implements active voltage control for low-voltage (LV) electrical distribution grids using inverter-based control of distributed energy resources (DERs). The framework supports both decentralized and distributed grid control strategies and adheres to the VDE-AR-N 4105 and EN50160 regulatory standards for providing ancillary services, such as reactive power control.

This repository is designed to the following:

• Reference Solution for Grid Control: Complies with VDE regulations and enables benchmarking against AI-driven control methods.
  • Reactive power control modes according to VDE-AR-N 4105:2018-11:
    • Q(V), Q(P), Fixed cos $ϕ$
  • Decentralized and Distributed DER grid control strategies in LV Grid:
    • Decentralized grid control: Each DER unit autonomously makes decisions at the local PCC, independent of grid constraints.
    • Distributed grid control: All DER units along the feeder control zones coordinate the decision-making process jointly.
• Detailed Grid Component Modelling:
  • Battery System Modelling: Captures charging and discharging dynamics, state-of-charge (SoC) constraints, and energy losses during storage operations.
  • Inverter Modelling: Simulates reactive and active power flows, efficiency losses, and compliance with regulatory voltage constraints.
  • Power Loss Analysis: Incorporates active power losses in distribution lines and transformers due to DER integration, based on physical equations.
• DER Integration and Simulation: Provides tools to simulate DER integration and their impact on autonomous reactive power control.
  • MPVBench: The benchmark dataset with real-time recordings of mini-photovoltaic (MPV) systems, also known as balcony power plants, including metadata of the corresponding used hardware.
    • Visit MPVBench GitHub repository for more information.
• Optional Monte Carlo Simulation: Supports uncertainty modeling and sensitivity analysis using stochastic parameters for more robust and flexible evaluation of control strategies.

This repository is structured in a few key folders:

├── helper                                <- Directory for helper scripts.
├── input                                 <- Directory for input data.
├── output                                <- Directory for output data.
├── test                                  <- Directory for test data.
├── yaml                                  <- Directory for configuration files.
├── __main__.py                           <- Main script.
├── monte_carlo_cs1.ipynb                 <- Notebook for Case Study 1.
├── monte_carlo_cs2.ipynb                 <- Notebook for Case Study 2.
├── monte_carlo_cs3.ipynb                 <- Notebook for Case Study 3.
├── execution_script_cs1.ipynb            <- Execution script for Case Study 1.
├── LICENSE.txt                           <- Licensing information.
├── README.md                             <- Primary README for developers.
├── exe_server_helper_cs1_job.sh          <- Script executes tasks defined by the master script to the SLURM scheduler for Case Study 1.
├── exe_server_master_cs1_job.sh          <- Script submits batch jobs for Case Study 1.
├── exe_server_master_cs2_job.sh          <- Script to the SLURM scheduler for Case Study 2.
├── exe_server_master_cs3_job.sh          <- Script to the SLURM scheduler for Case Study 3.
└── requirements.txt                      <- Dependencies required.

---

2. BACKGROUND

European power system hierarchy includes four primary levels: extra high voltage (220–400 kV), high voltage (HV, 60–220 kV), medium voltage (MV, 6–20 kV), and low voltage (LV, 230–400 V) grids, interconnected via transformers. The power generated from traditional large power plants and offshore wind farms feeds into the extra high and high voltage grids. Larger photovoltaic plants generally link to the medium voltage grid. Major industrial consumers directly connect to the high-voltage grid, while smaller industries connect to either the medium or low-voltage grid, depending on their specific power requirements.

Fig. 1: Illustration of the European Power System showing interactions and connections via transformers between the transmission grid, inclusive of extra high voltage, the distribution grid, comprising high-voltage (indicated in red), medium-voltage (shown in yellow), and LV (represented in black). Power absorption is depicted in red arrows, while power injection into the higher and lower levels is indicated with green arrows.

3. Problem Formulation: V&Q Control of DERs

The reactive power control indirectly influences the active power $p_{j}^{DER}$, with the separate inverters apparent powers for PV $s_{j}^{PV}$ and BES $s_{j}^{BES}$ systems determining the operating constraints:

$$ |q_{j}^{DER}| \leq \sqrt{(s_{j}^{PV})^2 - (p_{j}^{PV})^2} + \sqrt{(s_{j}^{BES})^2 - (p_{j}^{BES})^2} $$

The voltage drop $\Delta V_{ij} = V_{i} - V_{j}$ across a distribution line connecting nodes $i$ and $j$ can be approximated using the relationship:

$$ \Delta V_{ij} = \frac{r_{ij} \left( p_{j}^{Load} - p_{j}^{DER} \right) + x_{ij} \left( q_{j}^{Load} - q_{j}^{DER} \right)}{V_{j}} $$

where $V_{i}$ is the parent busbar's voltage, which serves as a reference value. The terms $r_{ij}$ and $x_{ij}$ denote the resistance and reactance, respectively, between buses $i$ and $j$ as depicted in Figure 2. The control strategies we've implemented for low voltage grids, as visualized below, are specifically depicted in Figure 3.

Fig. 2: Equivalent circuit in a radial electrical distribution system including Loads, PVs, BESs, and MPVs.

Fig. 3: Reactive power control for DERs, schematic representation of the methods based on the actual value (AV) for active power and the PCC voltage measurement, based on VDE-AR-N 4105 Grid Code.

---

4. Usage

Execute PIDE framework by using the main.py script as shown below.

1. Command Line

python __main__.py --pv_ctrl voltage_reactive_power_ctrl --storage_p_ctrl rbc_pvbes_distributed_sc_ctrl

python PIDE --pv_ctrl voltage_reactive_power_ctrl --storage_p_ctrl rbc_pvbes_distributed_sc_ctrl

2. Run main file only: main.py

Arguments:

Argument	Type	Default Value	Choices	Help
--benchmark	str	simbench	simbench, customised	Benchmark dataset: 'simbench', 'customised'.
--sb_code	str	1-LV-rural1--0-sw		Simbench code for the benchmark dataset.
--scenario	str	2-future	0-today, 1-near future, 2-future	Scenario options: '0-today', '1-near future', '2-future'.
--standard	str	vde	vde, customised	Standards: 'vde' (VDE-AR-N 4105) or 'customised' (e.g., IEEE Std 1547-2018).
--standard_mode	str	base	base, deadband, customised	Mode options: 'base', 'deadband', or 'customised'.
--timeseries_ctrl	str	control_module	control_module, test_mode, manual	Timeseries control mode.
--pv_ctrl	str	voltage_reactive_power_ctrl	datasource,voltage_reactive_power_ctrl,power_factor_active_power_ctrl,constant_power_factor_active_power_ctrl	Control mode for PV systems.
--storage_p_ctrl	str	rbc_pvbes_distributed_sc_ctrl	datasource,rbc_pvbes_decentralized_sc_ctrl,rbc_pvbes_distributed_sc_ctrl,rbc_pvbes_distributed_sc_dnc_ctrl,rbc_bes_dnc_ctrl	P-Control mode for storage systems.
--storage_q_ctrl	str	"voltage_reactive_power_ctrl"	datasource,voltage_reactive_power_ctrl,constant_power_factor_active_power_ctrl	Q-Control mode for storage systems.
--soc_initial	float	5.0	0.00-100.00	Initial state of charge (SoC) for storage systems.
--scaling_pv	float	1.0	0.-10.0	Scaling factor for PV capacity (0.0-10.0).
--scaling_load	float	1.0	0.-10.0	Scaling factor for load capacity (0.0-10.0).
--scaling_storage	float	1.0	0.-10.0	caling factor for storage capacity (0.0-10.0).
--time_mode	str	"selected"	selected, random, default	Time mode: 'selected', 'random', or 'default'.
--episode_start_hour	int	0	0-24	Start hour (0-24).
--episode_start_day	int	179	0-354	Start day (0-354).
--episode_start_min_interval	int	0	0-3	Starting interval (0-3).
--episode_limit	int	96	0-35136	Maximum number of time steps in an episode.
--max_iterations	int	35135	0-35136	Maximum number of iterations for the simulation.
--flag_monte_carlo	lambda	"false"	bool	Enable or disable Monte Carlo simulation mode.
--num_monte_carlo_runs	int	100		Number of Monte Carlo simulation runs.
--seed_value	int	42	0-100	Initial seed value for Monte Carlo simulations (range 0-100)..
--mpv_flag	lambda	"true"	bool	Enable or disable MPV analysis. false or true
--mpv_benchmark	str	"mpvbench"	simbench, mpvbench, customised	MPV benchmarking source: 'simbench', 'mpvbench', or 'customised'..
--mpv_scaling	float	0.60	0.-10.0	Scaling factor for MPV capacity (0.0-10.0).
--mpv_concentration_rate_percent	float	100.00	0.00-100.00	MPV concentration rate as a percentage (0.0-100.0).
--mpv_inverter_apparent_power_watt	int	800	600-1000	Maximum apparent power for MPV inverters (600-1000 W).
--mpv_solar_cell_capacity_watt	int	2000	800-2000	Maximum power for MPV solar cells (800-2000 W).

2. Alternatively, you can modify the arguments in the main.py file to change the default value parameters.

---

Configuration Files:

Additional configuration settings are available in the yaml directory:

• Local Environment (local)::

  • local/user_config_grid_manage_der.yaml: User-defined configuration settings.
  • local/default_config_grid_manage_der.yaml: Default configuration settings.

• HPC Environment (hpc)::

  • hpc/user_config_grid_manage_der.yaml: User-defined configuration settings.
    • hpc/default_config_grid_manage_der.yaml: Default configuration settings.

• 1. Open Configuration Files: Locate the configuration files under the local and hpc directories.

• 2. Edit Paths: Search for the path settings in each file and update them as follows:

  • local (on Windows): These settings are used on a local Windows machine.

    helper_path: C:\\Users\\<username>\\PIDE\\helper 
    input_data_path: C:\\Users\\<username>\\PIDE\\input\\data
    output_data_path: C:\\Users\\<username>\\PIDE\\output\\data
    output_test_path: C:\\Users\\<username>\\PIDE\\test

  • hpc (on Linux-based): These settings are intended for an HPC environment running on Linux.

    helper_path: /hkfs/home/haicore/iai/<username>/PIDE/helper
    input_data_path: /hkfs/home/haicore/iai/<username>/PIDE/input/data
    output_data_path: /hkfs/home/haicore/iai/<username>/PIDE/output/data
    output_test_path: /hkfs/home/haicore/iai/<username>/PIDE/test

---

4. Replicating Results:

To replicate the results from the paper, follow these steps:

1. Clone the repository:

   git clone https://github.com/KIT-IAI/PIDE.git

2. Navigate to the cloned directory:

   cd PIDE

3. Run the following command to install the packages:

   pip install -r requirements.txt

4. Run the provided bash scripts for each case study:

   • For Case Study 1:

     chmod +x ./exe_server_master_cs1_job.sh
     ./exe_server_master_cs1_job.sh

     • Calls exe_server_helper_cs1_job.sh to create SLURM jobs for sensitivity analysis.
     • Calls execution_script_cs1.ipynb to execute script for Case Study 1 on SLURM.
     • Open and execute the notebook monte_carlo_cs1.ipynb.

   • For Case Study 2:

     chmod +x ./exe_server_master_cs2_job.sh 
     ./exe_server_master_cs2_job.sh

     • Open and execute the notebook monte_carlo_cs2.ipynb.

   • For Case Study 3:

     chmod +x ./exe_server_master_cs3_job.sh
     sbatch ./exe_server_master_cs3_job.sh

     • Open and execute the notebook monte_carlo_cs3.ipynb.

---

License

This code is licensed under the MIT License.

Citation

If you use this framework in your research, please cite the following: [paper](under review):

• BibTeX format:

@article{Demirel2025,
  author    = {Demirel, Gökhan and Fernengel, Natascha and Grafenhorst, Simon and Förderer, Kevin and Hagenmeyer, Veit},
  title     = {{PIDE: Photovoltaic Integration Dynamics and Efficiency for Autonomous Control on Power Distribution Grids}},
  journal   = {Energy Informatics},
  year      = {2025},
  note      = {Under review}
}

Acknowledgment

This work was supported in part by:

• The Energy System Design (ESD) Project (Structure 37.12.02).
• The Helmholtz Association (HGF) Initiative and Networking Fund through Helmholtz AI.
• The HAICORE@KIT partition.

Contact Information

For any issues or any intention of cooperation, please feel free to contact me at goekhan.demirel@kit.edu.