A Physics-Regularized Gated Model with Patch-Based Temporal Encoding for Short-Term Probabilistic Wind Farm Power Forecasting

A physics-constrained temporal modeling framework for robust short-term wind power forecasting

Accurate short-term wind power prediction plays a critical role in ensuring stable grid operation, effective energy management, and the large-scale integration of renewable energy systems under highly variable wind conditions. Although data-driven and deep learning models have demonstrated promising forecasting capability, many existing approaches suffer from performance degradation during rapid wind fluctuations due to the lack of embedded physical constraints and the high computational complexity associated with recurrent architectures. To address these limitations, this article proposes a Physics-Guided Residual Temporal Convolutional Network (PG-ResTCN) for short-term wind power forecasting. The proposed framework integrates dilated temporal convolutional learning with residual connections to effectively capture multiscale temporal dependencies in wind power time series while avoiding the sequential computation of recurrent neural networks, thereby improving computational efficiency. Furthermore, a physics-based smoothness constraint is incorporated into the training loss function to enforce physically consistent power ramping behavior that reflects the inherent operational dynamics and inertia of wind turbines, reducing unrealistic fluctuations in predicted power outputs. The effectiveness of the proposed model is validated using a large real-world dataset containing approximately 140,161 hourly samples collected from five wind farm locations, including meteorological variables and corresponding turbine power outputs. Comprehensive experiments are conducted by comparing the proposed method with widely used benchmark models, including Support Vector Regression, Random Forest, Extreme Gradient Boosting, and Long Short-Term Memory networks. Results demonstrate that the proposed PG-ResTCN model achieves superior forecasting performance, obtaining a root mean square error of 0.0359, mean absolute error of 0.0296, and R 2 of 0.9759, outperforming all baseline models. The integration of physics-guided constraints with the temporal convolutional architecture significantly enhances prediction accuracy, stability, and generalization capability. In addition, the proposed framework maintains high computational efficiency and robustness, making it suitable for real-time wind energy forecasting applications, intelligent energy management systems, and microgrid power system operations.

Physics-guided temporal fusion transformer for probabilistic photovoltaic power forecasting under weather regime shifts

Accurate short-term photovoltaic (PV) power forecasting is critical for secure grid operation and economic dispatch, yet its performance is often degraded by non-stationary irradiance fluctuations induced by cloud transients and weather regime shifts. To address this challenge, this paper proposes a physics-guided temporal fusion architecture built on the Temporal Fusion Transformer (TFT) for multi-horizon probabilistic PV forecasting. The model fuses historical plant measurements, Numerical Weather Prediction (NWP) variables, and solar-geometry features, and introduces a cross-attention fusion module to align future meteorological drivers with the most relevant historical context for improved ramp responsiveness. Physical plausibility is promoted by soft physics-consistency regularization, including non-negativity, rated-capacity bounds, clear-sky envelope constraints, and ramp-rate penalties, while uncertainty is quantified via multi-quantile regression to produce calibrated prediction intervals. Experiments on real-world PV datasets across different seasons and weather conditions show that the proposed approach consistently improves deterministic accuracy (MAE and RMSE) and probabilistic performance (PICP and CRPS) over representative baselines, with particularly robust behavior under regime-shift and ramp events. Overall, combining attention-based temporal fusion with physics-guided learning provides a practical and scalable solution for reliable probabilistic PV power forecasting in highly variable atmospheric conditions.

Constraint-Guided Residual PatchTST With NWP Features for Multi-Horizon Probabilistic PV and Wind Power Forecasting

Accurate renewable power forecasting is essential for grid operation, reserve scheduling, and renewable integration. This paper proposes a constraint-guided residual forecasting framework that combines a domain-informed baseline with data-driven residual correction for multi-horizon probabilistic forecasting. A Gradient Boosting Regression Tree (GBRT) model is used as a classical residual benchmark, while a Residual PatchTST model captures temporal dependencies from numerical weather prediction features, engineered time variables, and site information. Final forecasts are reconstructed by adding the predicted residual to the baseline and enforcing nonnegative outputs within site-level capacity limits. Across PV sites, GBRT reduces mean RMSE from 0.0909 to 0.0798 and mean MAE from 0.0400 to 0.0358, while wind RMSE decreases from 0.2574 to 0.1479. The deep probabilistic model also delivers stable multi-horizon performance and useful uncertainty intervals. These results show that residual learning with constraint-guided reconstruction provides accurate and operationally meaningful renewable power forecasts.

Diffusion-enhanced representation learning with a parameterized Markov chain for short-term wind speed forecasting

Pangu-Weather Driven Dual-Path Network with Dynamic Gated Attention for Ultra-Short-Term Wind Power Forecasting

Results demonstrate that AI-weather-driven forecasting and dual-path dynamic fusion can provide a promising low-latency meteorological-input strategy for ultra-short-term wind power forecasting.

State-Aware Dynamic Retrieval and Persistence-Anchored Expert Fusion for Very Short-Term Wind Power Forecasting Without Numerical Weather Prediction

Very short-term wind power forecasting without future numerical weather prediction (NWP) must exploit historical observations while controlling operating-state heterogeneity, erroneous analog retrieval, and unstable expert corrections. This study proposes state-aware dynamic retrieval and persistence-anchored expert fusion (SADR-PAEF). A continuous operating-state representation supports training-only historical retrieval, while persistence, analog-trajectory, and neural residual experts are coordinated by a state-aware gate. Persistence-biased initialization, entropy-based contraction based only on gate-weight concentration, and validation-selected anchoring constrain excessive deviations from persistence; gating entropy is treated as a heuristic stabilization factor rather than a calibrated reliability measure. State-Adaptive Online Conformal Prediction (SAOCP) provides state-conditional online interval calibration. Across five runs on the 2024 main-site dataset, SADR-PAEF achieved an MSE of 44.0526 ± 0.2307 MW2, 3.22% lower than DLinear, with significant four-horizon-aggregated squared-error improvements over all seven baselines. Independent retraining on the 2019 site yielded consistent performance under another site distribution, which is interpreted as site-level replicability rather than cross-site transfer or zero-shot generalization. At 90% nominal coverage, SAOCP achieved the lowest overall Winkler score and the lowest Winkler score at all four horizons, although Rolling CP was marginally closer to nominal coverage at 15 min. Persistence retained a slight MAE advantage. The proposed framework therefore primarily reduces squared-error and large-deviation risk while providing adaptive interval calibration under NWP-unavailable conditions.