Source-linked AI summary
Non-Gaussian power grid frequency fluctuations characterized by Lévy-stable laws and superstatistics
Benjamin Schäfer, Christian Beck, Kazuyuki Aihara, Dirk Witthaut, Marc Timme
TL;DR
Power-grid frequency fluctuations are difficult to model because measurements across regions show substantial departures from Gaussianity, while prior approaches often simplify them as Gaussian. The paper analyzes multi-continent frequency time series with stochastic and superstatistical models, finding heavy tails, skewness, and trading-related fluctuations, while identifying damping and grid size as controls on fluctuation risk.
Problem
Existing studies often assume Gaussian noise, while systematic quantitative evidence connecting non-Gaussian frequency fluctuations to differently sized synchronous regions remains limited.
Method
The paper analyzes frequency time series across continents using an aggregated swing equation, generalized Fokker–Planck descriptions, and superstatistics for changing parameters.
Results
The analysis finds substantial non-Gaussian frequency fluctuations, including heavy tails and skewness, and identifies trading as a substantial source of fluctuations.
Takeaways & Limitations
Effective damping, inertia, and grid size are controlling factors for reducing fluctuation-induced risks, while longer-term modeling must account for changing Gaussian parameters or explicitly non-Gaussian distributions.
Takeaways & Limitations
The framework neglects spatial correlations, and questions involving correlated noise, microgrids, and separating damping from primary control require further data analysis.
Abstract
from arXiv · showhide
Multiple types of fluctuations impact the collective dynamics of power grids and thus challenge their robust operation. Fluctuations result from processes as different as dynamically changing demands, energy trading, and an increasing share of renewable power feed-in. Here we analyze principles underlying the dynamics and statistics of power grid frequency fluctuations. Considering frequency time series for a range of power grids, including grids in North America, Japan and Europe, we find a substantial deviation from Gaussianity best described as Lévy-stable and q-Gaussian distributions. We present a coarse framework to analytically characterize the impact of arbitrary noise distributions as well as a superstatistical approach which systematically interprets heavy tails and skewed distributions. We identify energy trading as a substantial contribution to today's frequency fluctuations and effective damping of the grid as a controlling factor enabling reduction of fluctuation risks, with enhanced effects for small power grids.
OBSERVING THE STATISTICS OF FREQUENCY FLUCTUATIONS
Across power grids, frequency fluctuations depart substantially from Gaussian behavior, showing heavy tails, skewness, and regular correlation peaks associated with electricity trading intervals.
- Grid-scale effects: Larger grids tend to have smaller frequency variance because their greater inertia reduces fluctuations.
- Distributional deviations: Frequency distributions show heavier tails than Gaussian predictions in Continental Europe, Nordic, Mallorca, and Japan, while Great Britain and Eastern Interconnection are substantially skewed.Lower-frequency deviations are more likely than higher-frequency deviations in the skewed grids.
- Distributional deviations: Lévy-stable and q-Gaussian distributions are the best-fitting distributions among those tested.The two families respectively accommodate intrinsically heavy-tailed or skewed input fluctuations and changing variance or mean over longer time scales.
- Temporal structure: 15-minute correlation peaks, especially pronounced in Great Britain and Central Europe, align with electricity-market trading intervals.At new trading intervals, production changes nearly instantaneously and the grid relaxes toward a new operational state.
- Temporal structure: Inverse correlation times have the same order of magnitude across grids, with γ defined as the inverse correlation time τ^-1.Japanese measurements were too sparse for autocorrelation estimation; the Japanese dataset had five-minute sampling.
STOCHASTIC MODEL OF POWER FLUCTUATIONS
The paper models bulk grid-frequency dynamics with an aggregated swing equation and Fokker–Planck framework that accepts different noise distributions. It shows how stable noise, damping, inertia, and grid size shape fluctuation risks.
- Model formulation: A Fokker–Planck equation provides an analytical frequency distribution once the noise distribution, damping, noise amplitudes, and total inertia are specified.
- Model formulation: The aggregated swing equation links bulk frequency dynamics to effective damping, inertia, coupling, and stochastic power fluctuations.The model uses coarse-grained generator or coherent-subgroup nodes and reduces their dynamics to bulk angular velocity.
- Non-Gaussian noise: Stable input noise preserves the output distribution’s skewness and stability parameter, changing only its scale under the linear stochastic model.This closure property applies to the aggregated swing equation and includes Gaussian noise as the αS → 2 limit.
- Risk controls and scaling: Increasing effective damping or inertia narrows the output distribution, whereas decreasing inertia increases the scale parameter and risk of large deviations.The stable-noise scaling is σS ∼ N^(αS−1)/αS, with the square-root law recovered only for αS = 2.
SUPERSTATISTICS
The paper interprets heavy tails and skewness either through non-Gaussian noise or through superstatistics, where slowly changing system parameters produce superimposed Gaussian distributions. Data-based time-scale and effective-friction analyses support this superstatistical interpretation across synchronous regions.
- Mechanism: Superstatistics explains observed heavy tails and skewness by superimposing Gaussian distributions with changing parameters.The damping γ and noise amplitude vary over time because grid operation, generation, demand, and connected power plants change.
- Noise amplitudes: Figure 5 estimates noise amplitudes under homogeneous inertia and equal node amplitudes, finding that they tend to increase with intermittent-renewable shares.Nordic-grid estimates have large uncertainty because of two different correlation time scales.
- Time scales: The long time scale T is selected from local kurtosis by the condition κ(∆t = T) = 3, with average skewness providing a parallel criterion.These criteria identify the interval over which an approximately Gaussian equilibrium distribution emerges.
- Time scales: The intrinsic short time scale is τ ≈200...550 s, whereas the long time scale is T ≈1...5 h across synchronous regions.T is at least one order of magnitude larger than τ, supporting the separation of time scales required by the approach.
- Consistency check: The effective friction γeff for Japanese 60Hz measurements is well described by a log-normal distribution, supporting the superstatistical model.The paper states that log-normal effective friction leads to approximate q-Gaussian frequency distributions.
DISCUSSION
The discussion presents non-Gaussian noise and superstatistics as complementary ways to model power-grid uncertainty, while linking fluctuation risks to grid parameters and operating scales. It also identifies unresolved questions about correlated noise, very small grids, and separating damping from primary control.
- DISCUSSION: The analysis identifies trading as a substantial source of power-grid frequency fluctuations and finds non-Gaussian behavior across synchronous regions.The paper analyzes frequency measurements from North America, Japan, and Europe using analytical stochastic methods.
- DISCUSSION: The derived framework predicts how power fluctuations affect grid-frequency distributions using the swing equation and generalized Fokker–Planck equations.The derivation neglects spatial correlations and focuses on bulk angular velocity.
- DISCUSSION: Increasing effective damping and inertia, alongside grid size, are identified as controlling factors for reducing fluctuation-induced risks.The paper connects modified damping through smart-grid control or generator droop control with a lower likelihood of large fluctuations.
- DISCUSSION: For uncertainty modeling, the paper proposes either Lévy-stable noise or superimposed Gaussian distributions, with Gaussian modeling supported mainly at time scales of one hour or below.Month- or year-scale studies must account for changing Gaussian means and variances or explicitly model non-Gaussian distributions.
- DISCUSSION: Fluctuation scaling and temporal correlations may support market design, mitigation timing, and damping estimates for isolated grids and microgrids.The discussion emphasizes that trading, correlation structure, and grid size connect statistical analysis to operation and design.
- DISCUSSION: Open questions concern correlated noise, scaling in larger collections of grids and microgrids, and disentangling damping from primary control.The authors call for more data and further analysis, particularly for microgrids.
Moments of the frequency distributions
The paper defines moments of grid-frequency distributions and uses skewness and kurtosis to quantify asymmetry and tail extremity.
- The n-th moment is computed from M measurements of the frequency variable f.The first moment is the mean, while the variance is the centralized second moment.
- Skewness β measures asymmetry around the mean, whereas kurtosis κ quantifies tail extremity.For Gaussian distributions, β = 0 and κ = 3.
- κCE = 4.0 ± 0.1 for the Continental European grid, exceeding the Gaussian benchmark κGauss = 3.The higher kurtosis indicates an increased likelihood of large deviations.
Normally distributed noise
The normally distributed noise model treats contributions from consumers, renewables, trading, and other sources as independent Gaussian variables whose sum remains Gaussian.
- The model represents each noise realization ξi as normally distributed.The notation N(0, 1) denotes mean 0 and standard deviation 1.
- The aggregate noise is formed by summing identically and independently distributed Gaussian variables.This sum represents combined contributions from consumers, renewables, trading, and other sources.
- The sum of independent Gaussian variables is distributed like a single normal distribution.This property motivates the Gaussian noise assumption in the model.
Superstatistics
The superstatistics analysis extracts local distributional and damping properties from frequency time series to characterize changing noise conditions over time.
- The analysis extracts local kurtosis and effective damping from frequency time series.It begins with a time series x(t) and its mean x̄.
- The long time scale T is selected where κ(∆t = T) = 3, indicating locally Gaussian noise without excess kurtosis.This criterion identifies the scale at which the local distribution is approximately Gaussian.
- The time-varying effective friction γeff is computed to characterize changing damping.The extracted damping is treated as a time-dependent quantity.
- γeff is expected to follow log-normal, χ2, or inverse χ2 distributions because these yield q-Gaussian distributions of x.These distributions provide alternative superstatistical descriptions of the fluctuating friction.
Data availability
Frequency recordings for several regions are publicly available, while Mallorca and Eastern Interconnection data were obtained directly from contributors.
- Frequency recordings for the CE, GB, Nordic, and Japanese regions are publicly available at their cited references.
- Mallorca data were provided by Eder Batista Tchawou Tchuisseu, and Eastern Interconnection data were provided by unnamed contributors.