Characterizing the load profile in power grids by Koopman mode decomposition of interconnected dynamics

Ali Tavasoli, Behnaz Moradijamei, Heman Shakeri

Abstract

Electricity load forecasting is crucial for effectively managing and optimizing power grids. Over the past few decades, various statistical and deep learning approaches have been used to develop load forecasting models. This paper presents an interpretable machine learning approach that identifies load dynamics using data-driven methods within an operator-theoretic framework. We represent the load data using the Koopman operator, which is inherent to the underlying dynamics. By computing the corresponding eigenfunctions, we decompose the load dynamics into coherent spatiotemporal patterns that are the most robust features of the dynamics. Each pattern evolves independently according to its single frequency, making its predictability based on linear dynamics. We emphasize that the load dynamics are constructed based on coherent spatiotemporal patterns that are intrinsic to the dynamics and are capable of encoding rich dynamical features at multiple time scales. These features are related to complex interactions over interconnected power grids and different exogenous effects. To implement the Koopman operator approach more efficiently, we cluster the load data using a modern kernel-based clustering approach and identify power stations with similar load patterns, particularly those with synchronized dynamics. We evaluate our approach using a large-scale dataset from a renewable electric power system within the continental European electricity system and show that the Koopman-based approach outperforms a deep learning (LSTM) architecture in terms of accuracy and computational efficiency. The code for this paper has been deposited in a GitHub repository, which can be accessed at the following address github.com/Shakeri-Lab/Power-Grids.

Introduction

Predicting electricity load is a high-stakes challenge for grid stability, especially as renewable energy integration increases system volatility. Traditional forecasting tools often face a trade-off between interpretability and accuracy. Statistical models like ARIMA are interpretable but often struggle with the complex, nonlinear nature of real-world power usage. Conversely, deep learning models like LSTMs offer high accuracy but function as "black boxes," requiring significant computational resources and offering little physical insight into their predictions. This paper introduces a framework that bridges this gap using Koopman operator theory to linearize nonlinear dynamics in a higher-dimensional space.

Methodology: The Koopman Framework

The core of the proposed approach is the Koopman Operator Theory. Instead of modeling the state of the grid directly, the method observes how functions of that state evolve. In this "observable" space, the dynamics become linear, allowing for simpler and more robust forecasting.

The proposed Koopman-based framework for load forecasting.

The technical pipeline involves several key steps as illustrated in Figure 1. First, Delay Embedding is used to reconstruct the system's manifold from historical lags. The method then employs Galerkin-Approximated Pseudospectra to find a finite-dimensional version of the infinite Koopman operator while remaining robust against spectral pollution. To handle the massive scale of the European grid (1,494 stations), the researchers use PHATE Clustering to group stations with synchronized dynamics. This allows for the construction of localized, highly efficient models. The process concludes with the computation of Dirichlet energies to order eigenfunctions by their complexity and the use of Nystrom extensions to generalize these eigenfunctions across the entire state space.

Spectral Insights and Physicality

A significant advantage of the Koopman approach is its inherent interpretability. The computed eigenfunctions correspond to real-world physical cycles rather than abstract weights in a neural network.

Spectral analysis and corresponding Koopman eigenfunctions.

As shown in Figure 2, the researchers identified dominant frequencies that scale precisely with human and environmental cycles: seasonal weather (monthly), work-week patterns (weekly), and solar/human activity cycles (daily). The framework also allows for a unique "noise decomposition," distinguishing between innovation noise (actual random fluctuations) and modal noise (errors caused by truncating the number of eigenfunctions). This provides engineers with actionable feedback on model performance.

Performance and Efficiency

The study utilized the "Re-europe" dataset, a benchmark of the continental European electricity system featuring 1,494 transmission buses and 2,156 lines with a 1-hour resolution. Key results include:

  • Superior Accuracy: The Koopman-based approach consistently achieved lower Root Mean Squared Error (RMSE) than a separately optimized LSTM architecture across various clusters and seasons.
  • Massive Efficiency: The proposed method was 36 times faster to compute than the LSTM, significantly reducing the computational burden for grid operators.
  • Dimensionality Reduction: The authors successfully reconstructed the complex load profiles of all 1,494 stations using only 100 Koopman eigenfunctions, proving that the underlying dynamics are governed by a relatively small set of coherent patterns.

Conclusion

By representing nonlinear power grid dynamics as a linear system in a higher-dimensional space, the Koopman-based framework provides a powerful alternative to black-box deep learning. It offers superior accuracy, massive computational speedups, and clear physical insights into the grid's behavior. While challenges remain regarding sudden exogenous shocks and the memory costs of high-delay embeddings, this work marks a significant step toward more reliable and interpretable power grid management.

Summary updated

Abstract

Electricity load forecasting is crucial for effectively managing and optimizing power grids. Over the past few decades, various statistical and deep learning approaches have been used to develop load forecasting models. This paper presents an interpretable machine learning approach that identifies load dynamics using data-driven methods within an operator-theoretic framework. We represent the load data using the Koopman operator, which is inherent to the underlying dynamics. By computing the corresponding eigenfunctions, we decompose the load dynamics into coherent spatiotemporal patterns that are the most robust features of the dynamics. Each pattern evolves independently according to its single frequency, making its predictability based on linear dynamics. We emphasize that the load dynamics are constructed based on coherent spatiotemporal patterns that are intrinsic to the dynamics and are capable of encoding rich dynamical features at multiple time scales. These features are related to complex interactions over interconnected power grids and different exogenous effects. To implement the Koopman operator approach more efficiently, we cluster the load data using a modern kernel-based clustering approach and identify power stations with similar load patterns, particularly those with synchronized dynamics. We evaluate our approach using a large-scale dataset from a renewable electric power system within the continental European electricity system and show that the Koopman-based approach outperforms a deep learning (LSTM) architecture in terms of accuracy and computational efficiency. The code for this paper has been deposited in a GitHub repository, which can be accessed at the following address github.com/Shakeri-Lab/Power-Grids.
BibTeX
@misc{tavasoli2023characterizing,
  author = {Ali Tavasoli and Behnaz Moradijamei and Heman Shakeri},
  title = {{Characterizing the load profile in power grids by Koopman mode decomposition of interconnected dynamics}},
  year = {2023},
  eprint = {2304.07832},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2304.07832},
}
Figure