Towards turbine-location-aware multi-decadal wind power predictions with CMIP6

Nina Effenberger, Nicole LudwigarXiv 2024 · DOI · arXiv · OpenReview

TL;DR

This paper introduces a novel approach using Gaussian Processes to predict multi-decadal wind power generation in Germany, crucially incorporating turbine locations. The findings suggest that wind energy will remain a reliable source, with location-aware predictions offering greater accuracy and highlighting regional differences in future uncertainty.

What Problem Does This Paper Address?

Accurately predicting future wind power generation is critical for effective energy planning, especially in the context of climate change. However, global climate models (GCMs) used for these projections have coarse spatial resolutions, making it difficult to capture localized wind patterns. Furthermore, these models inherently carry uncertainty due to unpredictable human behavior and the complexity of climate systems. This paper aims to bridge this gap by developing a method that leverages GCM data but refines predictions by considering the specific locations of wind turbines, thereby improving accuracy and providing more granular insights into future wind energy potential.

Methods

The core of this research lies in the application of Gaussian Processes (GPs) for predicting wind power. GPs are a powerful statistical tool that can model complex relationships and quantify uncertainty. The authors use GPs to:

  • Estimate wind speeds at turbine locations: Instead of relying solely on gridded GCM data, GPs are used to interpolate and predict wind speeds at the precise locations of existing wind turbines. This is a key innovation, moving beyond coarse grid-level predictions.
  • Extrapolate wind speeds to hub height: Wind speed measurements are typically taken at a standard height (e.g., 10m), but turbines operate at higher hub heights. A wind profile power law is used to extrapolate these speeds to the relevant turbine hub height.
  • Compute wind power output: The predicted wind speeds at hub height are then fed into turbine-specific power curves to estimate the power output.
  • Location-aware vs. Non-location-aware predictions: The study explicitly compares predictions that incorporate turbine locations with those that do not, highlighting the added value of the former.
  • Hyperparameter Optimization: GPs have hyperparameters that are optimized to best fit the data, ensuring the model's performance.

Datasets

The study utilizes the following datasets:

  • ERA5 Reanalysis Data: A high-resolution gridded reanalysis dataset used for evaluating historical wind speed predictions.
  • CMIP6 Climate Model Data (MPI-ESM1.2-HR): This dataset provides future climate projections under various Shared Socioeconomic Pathways (SSPs), including SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5.
  • Turbine Location Data: A dataset containing the geographical locations and characteristics of wind turbines in Germany, provided by Manske and Schmiedt [30].
  • Actual Power Generation Data: Transmission System Operator (TSO) data and the SMARD database are used to represent actual wind power generation in Germany for validation.

Key Insights and Ideas

  • Turbine Location is Crucial: The most significant finding is that accounting for the specific locations of wind turbines dramatically improves the accuracy of multi-decadal wind power predictions compared to methods that only use gridded data. This emphasizes that "where" the turbines are matters more than just "how much" wind is projected.
  • Wind Energy Remains Reliable: Despite climate change, the study suggests that wind energy will likely remain a reliable power source in Germany through 2050, with only minor changes in yearly power generation projected under most scenarios.
  • SSP2-4.5 and SSP3-7.0 as Realistic Scenarios: The in-between climate scenarios SSP2-4.5 and SSP3-7.0 show projections that closely align with actual observed wind power generation, suggesting these might be more representative of future conditions than extreme scenarios.
  • Spatial Uncertainty Varies: There's a notable difference in prediction uncertainty across Germany, with coastal northern regions exhibiting higher uncertainty than the mountainous south. This has direct implications for future wind farm development.

Key Results

  • Improved Accuracy: Location-aware predictions showed significantly better agreement with actual power generation compared to non-location-aware predictions. For example, in historical validation, location-aware predictions with ERA5 overestimated by only 5.02%, while non-location-aware predictions underestimated by 9.49%.
  • Minor Future Changes: For Germany, predictions up to 2050 reveal only minor changes in yearly wind power generation across the analyzed CMIP6 scenarios.
  • Regional Uncertainty: The study highlights higher uncertainty in wind power predictions for Germany's coastal North compared to its South, suggesting that the South might be a more reliable region for future wind power expansion.
  • Scenario Alignment: For the period 2015-2023, wind power projections from SSP2-4.5 and SSP3-7.0 closely matched actual power generation.

Limitations and Future Work

  • Data Confidentiality: The authors acknowledge that much of the actual power generation data is confidential, which limits the ability to fully account for factors like curtailment and technical improvements in turbines.
  • Hub Height Data: Direct measurements of wind speed at hub height are not readily available, necessitating extrapolation using power laws, which introduces its own uncertainties.
  • Location-Specific vs. Location-Aware: While the method is "location-aware" (considering turbine locations), it is not fully "location-specific" (predicting for every potential new location).
  • Future Work: The authors suggest several avenues for future research:
  • Investigating other potential turbine scenarios and expanding storage capacities.
  • Further exploiting the uncertainty quantification capabilities of GPs, potentially with larger ensembles.
  • Validating the methodology for larger regions, which is challenging due to the lack of a common wind turbine installation database.
  • Exploring physics-informed GP kernels.
  • Improving vertical extrapolation and learning probabilistic power curves.

Summary updated

Abstract

Climate change will impact wind and therefore wind power generation with largely unknown effect and magnitude. Climate models can provide insights and should be used for long-term power planning. In this work we use Gaussian processes to predict power output given wind speeds from a global climate model and compare the aggregated predictions to actual power generation. Analyzing past climate model data supports the use of CMIP6 climate model data for multi-decadal wind power predictions and highlights the importance of being location-aware. Our predictions up to 2050 reveal only minor changes in yearly wind power generation. We find that wind power projections of the two in-between climate scenarios SSP2-4.5 and SSP3-7.0 closely align with actual wind power generation between 2015 and 2023. Our analysis also reveals larger uncertainty associated with Germany's coastal areas in the North as compared to Germany's South, motivating wind power expansion in regions where future wind is likely more reliable. Overall, our results indicate that wind energy will likely remain a reliable energy source in the future.
BibTeX
@misc{effenberger2024towards,
  author = {Nina Effenberger and Nicole Ludwig},
  title = {{Towards turbine-location-aware multi-decadal wind power predictions with CMIP6}},
  year = {2024},
  eprint = {2408.14889},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2408.14889},
}
Figure