On autoregressive deep learning models for day-ahead wind power forecasting with irregular shutdowns due to redispatching

TL;DR

This work tackles day-ahead wind power forecasting for redispatch planning in the presence of irregular shutdowns. It systematically compares autoregressive deep-learning models (DeepAR, NHiTS, TFT) with wind-power-curve modeling approaches (OEM curve, AutoWP, and ML-based curves) and assesses shutdown-handling strategies. The findings show that WP-curve modeling, particularly AutoWP, generally achieves lower forecasting errors and greater scalability than autoregressive DL models, which require extensive data cleaning and high computational effort. These results have practical implications for deploying scalable, reliable forecasts across distributed onshore turbines in smart-grid operations, and point to future work with larger, labeled datasets and potential hybrid strategies.

Abstract

Renewable energies and their operation are becoming increasingly vital for the stability of electrical power grids since conventional power plants are progressively being displaced, and their contribution to redispatch interventions is thereby diminishing. In order to consider renewable energies like Wind Power (WP) for such interventions as a substitute, day-ahead forecasts are necessary to communicate their availability for redispatch planning. In this context, automated and scalable forecasting models are required for the deployment to thousands of locally-distributed onshore WP turbines. Furthermore, the irregular interventions into the WP generation capabilities due to redispatch shutdowns pose challenges in the design and operation of WP forecasting models. Since state-of-the-art forecasting methods consider past WP generation values alongside day-ahead weather forecasts, redispatch shutdowns may impact the forecast. Therefore, the present paper highlights these challenges and analyzes state-of-the-art forecasting methods on data sets with both regular and irregular shutdowns. Specifically, we compare the forecasting accuracy of three autoregressive Deep Learning (DL) methods to methods based on WP curve modeling. Interestingly, the latter achieve lower forecasting errors, have fewer requirements for data cleaning during modeling and operation while being computationally more efficient, suggesting their advantages in practical applications.

Figures (7)

• Figure 1: Using the WP curve to make forecasts requires the wind speed forecast to be corrected to the curve's reference height.
• Figure 2: The three steps of AutoWP's automated design exemplified for the real-/world WP turbine no. 1.
• Figure 3: Power generation $y$ and covariates available in the forecasting model's future and past horizon. $x_\text{sd}$ labels identified shutdowns in the past, $\hat{y}_\text{wpc}$ is the turbine's theoretical power generation according to the OEM WP curve, $\hat{v}_\text{eff}$ is the wind speed forecast at hub height, and the features $x_\text{s, 365}$, $x_\text{c, 365}$, $x_\text{s, 1440}$, and $x_\text{c, 1440}$ to encode temporally recurring shutdowns patterns. Dashed lines visualize unavailable future or unused past periods, and the horizontal line marks the forecast origin.
• Figure 5: Identified WP turbine shutdowns with rule-/based filtering Meisenbacher2024d. Black fields represent data points at a measured wind speed greater than the cut-/in speed with a power generation lower than the cut-/in power. The shutdowns for WP turbine no. 1 amount to $49%$ and $20%$ for no. 2.
• Figure 6: Data pre-/processing to identify abnormal operational states Meisenbacher2024d. Rule-/based filtering identifies turbine shutdowns and LOF-/based filtering identifies turbine stop-/to-/operation transitions and vice versa. The samples for which the wind speed at hub height is below the cut-/in speed $v_\text{cut-in}=2.5m\per s$ amount to $16.7%$, while the cut-/out speed $v_\text{cut-out}=25m\per s$ at which the WP turbine would be shut down for safety reasons is not reached.