Identification of Power Systems with Droop-Controlled Units Using Neural Ordinary Differential Equations

Hannes M. H. Wolf,Christian A. Hans

2024-11-13

Abstract:In future power systems, the detailed structure and dynamics may not always be fully known. This is due to an increasing number of distributed energy resources, such as photovoltaic generators, battery storage systems, heat pumps and electric vehicles, as well as a shift towards active distribution grids. Obtaining physically-based models for simulation and control synthesis can therefore become challenging. Differential equations, where the right-hand side is represented by a neural network, i.e., neural ordinary differential equations (NODEs), have a great potential to serve as a data-driven black-box model to overcome this challenge. This paper explores their use in identifying the dynamics of droop-controlled grid-forming units based on inputs and state measurements. In numerical studies, various NODE structures used with different numerical solvers are trained and evaluated. Moreover, they are compared to the sparse identification of nonlinear dynamics (SINDy) method. The results demonstrate that even though SINDy yields more accurate models, NODEs achieve good prediction performance without prior knowledge about the system's nonlinearities which SINDy requires to work best.

Systems and Control

What problem does this paper attempt to address?

The paper attempts to address the issue that in future power systems, due to the increase in distributed energy resources (such as photovoltaic generators, battery storage systems, heat pumps, and electric vehicles) and the shift towards active distribution networks, the detailed structure and dynamics of the system may not be fully known. This makes physics-based modeling difficult. To this end, the paper explores the use of Neural Ordinary Differential Equations (NODEs) to identify the dynamics of droop-controlled grid-forming units based on input and state measurements to overcome this challenge.

Specifically, the main objectives of the paper include:
1. **Applying NODEs to identify the dynamics of the entire power system containing droop-controlled grid-forming units**: This includes the dynamics of the nodes and their nonlinear coupling through the electrical network. By learning the dynamics from node measurement data, it is possible to predict the voltage and frequency dynamics of the system.
2. **Learning models that can be used for control synthesis and closed-loop simulation**: Unlike studies that only predict trajectories under fault scenarios, this paper aims to learn input-output system models.
3. **Comparing the performance of NODEs and SINDy (Sparse Identification of Nonlinear Dynamics)**: Although SINDy is more accurate, NODEs can achieve good predictive performance without requiring prior knowledge of system nonlinearities.
4. **Conducting extensive hyperparameter optimization**: The study examines the impact of network size and activation functions on model accuracy, concluding that smaller networks and continuous activation functions are more advantageous for the task.
5. **Exploring the impact of different integration schemes on NODEs**: The results show that using higher-order solvers does not necessarily lead to more accurate predictions.

Through these objectives, the paper aims to demonstrate the potential of NODEs in complex power systems and provide theoretical and technical support for their practical application.