Abstract
This deliverable provides an overview of the methodologies and results developed within Task 3.2 of the SUDOCO project, which aims to create control-friendly economic models for offshore wind farms. The primary focus is on developing forecasting models for electricity prices and weather conditions, which are essential for optimizing wind farm operations. These forecasts are used to inform wind power plant control strategies, enabling wind farms to adapt dynamically to changing market conditions and atmospheric factors.
The report details two major forecasting approaches: short-term and long-term models. Short-term models predict electricity prices and weather conditions for immediate use in wind farm control, while long-term models aim to project future electricity prices based on climate scenarios and energy system simulations. These techniques are applied to three cases studies: the Hollandse Kust Noord (HKN) wind farm (located near the Netherlands), the Danish Energy Island project, and a Floating Wind Farm near the coast of Portugal.
The newly proposed ”Cost of Value of Energy” (COVE) metric is introduced as an alternative to the traditional Levelized Cost of Energy (LCOE), offering a more comprehensive financial assessment by accounting for fluctuating electricity market values. Several machine learning and statistical methods are evaluated for their effectiveness in forecasting to support COVE-based optimization, including XGBoost, LightGBM, neural networks, and more traditional approaches like autoregressive integrated moving average and Ridge Regression. The results from these models are validated using cross-validation techniques, ensuring robust performance.
Key findings from this report include:
• Introduction of the COVE metric, which aligns more closely with market-driven energy systems than the traditional LCOE.
• The identification of the best-performing forecasting models for short-term predictions of electricity prices, direct and global solar radiation, wind speed, and wind direction for three case studies.
• Demonstration that incorporating weather patterns significantly improves the accuracy of long-term electricity price forecasts.
The report details two major forecasting approaches: short-term and long-term models. Short-term models predict electricity prices and weather conditions for immediate use in wind farm control, while long-term models aim to project future electricity prices based on climate scenarios and energy system simulations. These techniques are applied to three cases studies: the Hollandse Kust Noord (HKN) wind farm (located near the Netherlands), the Danish Energy Island project, and a Floating Wind Farm near the coast of Portugal.
The newly proposed ”Cost of Value of Energy” (COVE) metric is introduced as an alternative to the traditional Levelized Cost of Energy (LCOE), offering a more comprehensive financial assessment by accounting for fluctuating electricity market values. Several machine learning and statistical methods are evaluated for their effectiveness in forecasting to support COVE-based optimization, including XGBoost, LightGBM, neural networks, and more traditional approaches like autoregressive integrated moving average and Ridge Regression. The results from these models are validated using cross-validation techniques, ensuring robust performance.
Key findings from this report include:
• Introduction of the COVE metric, which aligns more closely with market-driven energy systems than the traditional LCOE.
• The identification of the best-performing forecasting models for short-term predictions of electricity prices, direct and global solar radiation, wind speed, and wind direction for three case studies.
• Demonstration that incorporating weather patterns significantly improves the accuracy of long-term electricity price forecasts.
| Original language | English |
|---|
| Publisher | European Union |
|---|---|
| Number of pages | 27 |
| Publication status | Published - 2024 |
