Temporally partitioned wind power forecasting with validation-calibrated residual diagnostics for PHM-oriented interpretation

Introduction Accurate wind-power forecasting is critical for the reliable integration of renewable energy into power grids, but reproducible evaluation requires explicit temporal partitioning and clearly defined feature-availability assumptions. This study presents a temporally partitioned, forecasting-first workflow utilizing the DaKS/Kassel synthetic wind-power dataset. Methods The canonical tabular benchmark establishes a target-time-aligned baseline by combining DaKS day-ahead numerical weather prediction covariates with recent observed-power history constructed relative to target valid time. Preprocessing parameters are fitted on the training partition, model and warning-threshold selection use validation data, and the test partition is reserved for final evaluation. For comparison, controlled subset experiments on four parks evaluated several advanced neural and sequential architectures, including a multilayer perceptron (MLP), a gated recurrent unit (GRU), a long short-term memory (LSTM) network, a temporal convolutional network (TCN), and PatchTST-Lite (Patch Time Series Transformer). Furthermore, we developed a validation-calibrated residual diagnostic layer that analyzes prediction errors across spatial, temporal, and operating-regime dimensions. Results Within the canonical comparable full-dataset benchmark, XGBoost achieved the lowest test mean absolute error (MAE) of 0.0657, a root mean squared error (RMSE) of 0.1134, and a coefficient of determination (R²) of 0.860. The strong-model diagnostic run achieved a test MAE of 0.0662 while retaining structured candidate warning episodes. Discussion These residual patterns are interpreted as structured deviations in forecast behavior rather than as confirmed physical equipment faults. This framework provides a reproducible forecasting and residual-screening foundation for future research in wind-energy digital twins and prognostics and health management (PHM).

Authors

Institutions

Publication Details

Journal
Frontiers in Energy Research
Published
2026-11-17
DOI
https://doi.org/10.3389/fenrg.2026.1948217
Primary Topic
Energy Load and Power Forecasting
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Temporally partitioned wind power forecasting with validation-calibrated residual diagnostics for PHM-oriented interpretation

Kyriaki Saiti, Christos Emexidis, Dionysios Alexopoulos
4 links
Frontiers in Energy Research
Energy Load and Power Forecasting
article

Temporally partitioned wind power forecasting with validation-calibrated residual diagnostics for PHM-oriented interpretation

Kyriaki Saiti, Christos Emexidis, Dionysios Alexopoulos
article en

Abstract

Introduction Accurate wind-power forecasting is critical for the reliable integration of renewable energy into power grids, but reproducible evaluation requires explicit temporal partitioning and clearly defined feature-availability assumptions. This study presents a temporally partitioned, forecasting-first workflow utilizing the DaKS/Kassel synthetic wind-power dataset. Methods The canonical tabular benchmark establishes a target-time-aligned baseline by combining DaKS day-ahead numerical weather prediction covariates with recent observed-power history constructed relative to target valid time. Preprocessing parameters are fitted on the training partition, model and warning-threshold selection use validation data, and the test partition is reserved for final evaluation. For comparison, controlled subset experiments on four parks evaluated several advanced neural and sequential architectures, including a multilayer perceptron (MLP), a gated recurrent unit (GRU), a long short-term memory (LSTM) network, a temporal convolutional network (TCN), and PatchTST-Lite (Patch Time Series Transformer). Furthermore, we developed a validation-calibrated residual diagnostic layer that analyzes prediction errors across spatial, temporal, and operating-regime dimensions. Results Within the canonical comparable full-dataset benchmark, XGBoost achieved the lowest test mean absolute error (MAE) of 0.0657, a root mean squared error (RMSE) of 0.1134, and a coefficient of determination (R²) of 0.860. The strong-model diagnostic run achieved a test MAE of 0.0662 while retaining structured candidate warning episodes. Discussion These residual patterns are interpreted as structured deviations in forecast behavior rather than as confirmed physical equipment faults. This framework provides a reproducible forecasting and residual-screening foundation for future research in wind-energy digital twins and prognostics and health management (PHM).

Frontiers in Energy ResearchVol. 14
New York College (GR)
Affordable and clean energy, Climate action
Energy Load and Power Forecasting
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon