The transition to sustainable energy positions wind power as a key renewable solution. As demand grows, wind turbines are deployed across diverse terrains. However, wind’s stochastic nature and environmental variability complicate power forecasting, affecting grid stability. The study leverages data-driven techniques to enhance wind power forecasting using high-resolution SCADA system time-series data. Key operational parameters include wind speed, rotor speed, generator speed, nacelle orientation, ambient temperature and power output. A comparative analysis evaluates traditional machine learning models—Linear Regression, Decision Trees, Random Forests, Gradient Boosting and Support Vector Machines—against deep learning models like Long Short-Term Memory (LSTM) networks and a novel Recurrent Neural Network (RNN) architecture. The core contribution is an optimized Bidirectional LSTM-RNN model with permutation layers and attention. These layers capture long-range dependencies and nonlinear interactions in wind data. The structure improves long-range dependency capture and nonlinear interaction modeling. Bidirectionality enables learning from both past and future time steps, while attention mechanisms highlight critical temporal features. Experimental results demonstrate the proposed model’s superior performance, achieving a Mean Absolute Error (MAE) of 0.0994 and Root Mean Square Error (RMSE) of 0.1390, significantly outperforming traditional models (e.g., Random Forest: MAE 86.44, RMSE 220.30) and basic LSTM models (MAE 14.48, RMSE 15.27). Robust cross-validation confirms its ability to generalize across different temporal segments. Feature importance analysis improves interpretability, supporting informed decision-making in wind farm operations. The framework is scalable, modular and well-suited for real-time forecasting applications. The work presents a reliable deep learning model for wind power forecasting, enabling intelligent, data-driven energy management in modern power systems.

Time-frequency feature analysis: A KAN-DLinear cooperative framework for multi-confidence interval wind power forecasting

ABSTRACT Accurate short-term wind power forecasting is essential for power system stability, peak-shaving, and renewable energy accommodation. However, achieving high-precision forecasting remains a significant challenge due to the inherent strong variability of wind energy and the complex coupling relationships among meteorological factors. To address this, this paper proposes a short-term wind power forecasting method that accounts for the coupling effects of meteorological factors. First, the mutual information method is employed to quantify the nonlinear dependencies between power output and meteorological variables, revealing latent coupling effects that traditional linear correlation analysis fails to capture, thereby facilitating the selection of key features. Second, an ensemble model incorporating TimeDistill is proposed for prediction. This model establishes a dual-teacher-student distillation architecture, utilizing BiLSTM and BiGRU as teacher networks to extract long-term dependencies and short-term dynamic features, respectively, guiding a Conv-MLP student network in learning complex temporal representations. Finally, the superiority of the proposed method is validated using measured data from a wind farm in Inner Mongolia. Experimental results demonstrate that, compared with LSTM and Transformer benchmark models, the proposed method reduces the Root Mean Square Error (RMSE) by 9.4% and 4.0%, respectively, providing an effective solution for real-time power dispatching.

ABSTRACT Ultra short-term wind power forecasts are core to grid dispatch for load coordination, power balancing, and secure operation. Forecast errors disrupt schedules, undermine system balance, threaten grid stability, and incur wind farm penalties. Thus, improving accuracy is critical to reducing operational risks and boosting renewable energy integration. To address key limitations of existing methods, insufficient nonlinear interaction modeling between trend and seasonal components, weak informative feature identification, and mismatched loss functions, this paper proposes an enhanced DLinear-based approach for ultra short-term wind power forecasting. The method innovatively designs a Decomposition-Interaction framework. First, a moving average kernel decomposes the original time series into trend and seasonal components; then, an interactive learning module captures their dynamic interdependencies. Subsequently, a dual Channel-Token attention mechanism is constructed and specifically applied to the seasonally refined component, leveraging its rich nonlinearity and pronounced periodicity, to simultaneously align global dependencies across multidimensional features and enhance local critical patterns, thereby improving feature representation effectiveness. Furthermore, a hybrid loss function combining Mean Squared Error (MSE) and Mean Absolute Error (MAE) with equal weighting is adopted to balance strong penalty on large errors with overall prediction robustness. Experiments conducted on both the Yalova dataset and an eastern Chinese wind farm dataset demonstrate that the proposed method achieves superior performance in ultra-short-term forecasting tasks. Specifically, it reduces MSE by 1.0%-3.9% on average compared to seven baseline approaches, offering wind farms a highly accurate and robust forecasting solution.

Accurate and reliable forecasting of wind power generation is a cornerstone for ensuring the operational flexibility, economic efficiency, and grid stability of modern power systems with high renewable penetration. While recent advances in deep learning have improved point and probabilistic forecasting, most existing methods still emphasize error minimization at isolated horizons or locations, neglecting the spatiotemporal dependencies, uncertainty calibration, and temporal coherence that are critical in real-world decision-making contexts. In this study, we propose a novel adaptive spatiotemporal graph neural network (ST-GNN) framework for multi-horizon probabilistic wind power forecasting, designed to jointly optimize deterministic accuracy, distributional sharpness, and temporal stability. The framework integrates site-specific meteorological inputs and SCADA data into a dynamically evolving graph structure, where edges are adaptively reweighted to capture changing inter-site correlations driven by wind climatology. A unified multi-objective loss function is formulated, combining mean error minimization with probabilistic metrics such as the Continuous Ranked Probability Score (CRPS) and a temporal smoothness regularizer that mitigates spurious fluctuations across consecutive horizons. Comprehensive case studies are conducted using two years of high-resolution data from a wind farm cluster consisting of twelve geographically diverse sites. Results demonstrate that the proposed model achieves up to 13% reduction in RMSE at 1-4 hour horizons and maintains improvements above 5% at 24-hour horizons when compared with state-of-the-art baselines. Probabilistic evaluations indicate a reduction in calibration error, with empirical coverage deviations generally within 3% across central quantiles and uncertainty estimates that adapt to different meteorological regimes. Moreover, the temporal rank-order consistency of predictions remains significantly higher than existing models, reducing dispatch-related instabilities and improving multi-site coordination. These findings suggest that the integration of spatiotemporal graph structures with probabilistic training objectives can provide a flexible and potentially generalizable paradigm.

Benchmarking of wind power based on data predictability and economic analysis using hybrid NARX algorithm

Enhancing wind power forecasting accuracy with hybrid temporal convolutional networks and adaptive stacked ensemble learning

Deep spatio-temporal modeling for improved wind and solar power forecasts in power grid integration

In Internet of Things (IoT)-enabled energy systems, accurate wind power forecasting is essential for intelligent scheduling and grid stability. However, meteorological and power time series often exhibit time-varying distributions and irregular regime shifts. This complexity complicates stable pattern learning and may cause models to overemphasize repeated local patterns. To address this issue, we propose a Cluster-Driven Block-Flip Transformer with Dynamic Positional Learning (CBFT) for wind power forecasting in IoT-based energy applications. CBFT includes two components. First, a block-based clustering and flipping module splits the input into fixed-length blocks and clusters them by statistical similarity. It then applies controlled randomized flipping to selected blocks as structured regularization to reduce order-sensitive dependencies. Second, a dynamic positional learning module uses multi-layer perceptron (MLP) embeddings to learn adaptive relative positional representations among blocks. This design remains effective under block-level order perturbations. Comprehensive evaluations on multiple real-world wind power and electricity datasets show that CBFT achieves lower MAE and MSE than state-of-the-art Transformer-based models.

A modular based framework for multi-step spatio-temporal wind power forecasting

Evaluating the impact of Numerical Weather Prediction variables on wind power forecasting: A case study of the Alpha Ventus offshore wind farm

Evaluation of wind power forecast uncertainty at German transmission grid buses: spatial patterns and ensemble spread reliability

Spatio-temporal Transformer-based modeling of multi-turbine wake effects for wind farm power forecasting

Conformalized Uncertainty Quantification for Joint Solar–Wind Power Forecasting in Reliable Grid Integration

GWO-DAGRU: A hybrid deep learning framework with metaheuristic feature selection and self-weighted context GRU for short-term wind power forecast

A Methodology for Forecasting Wind Power Generation in Extreme Weather Conditions and Evaluating Associated Risks

Against the background of accelerated green energy development and the deep integration of intelligent sensing technologies, wind power forecasting is evolving toward a multimodal sensor collaborative perception paradigm within nonlinear multi-source integrated energy systems. To address the limitations of conventional methods, including the lack of dynamic importance modeling and constrained stability under complex wind conditions, a forecasting framework based on multimodal sensor importance perception is proposed. This study emphasizes the framework’s role in decoding the complex nonlinear dependencies between atmospheric drivers and turbine responses. Through a multimodal feature encoding architecture, unified temporal representations of meteorological environments and turbine operational states are established. A sensor-importance-aware attention mechanism and a cross-modal relational modeling strategy are introduced to adaptively allocate contributions under varying contexts. Furthermore, prediction compensation and uncertainty characterization modules are integrated to enhance robustness. Systematic experiments on real-world multi-source data validate the method. Overall performance comparisons demonstrate that MAE, RMSE, and MAPE reach 30.48, 42.37, and 9.16 percent, respectively, with the coefficient of determination achieving 0.957, significantly outperforming the Transformer baseline. In multi-horizon tasks, the model exhibits superior error accumulation suppression, with twelve-step forecasting errors remaining at 41.27 and 56.48. These findings reveal that the framework captures the context-dependent nonlinear mapping of energy systems, providing effective technical support for green energy dispatch and intelligent sensing applications.

Short-term wind power forecasting is considered a critical challenge in smart grid management due to the nonlinear, unstable, and multi-scale noise characteristics of wind signals. Although recent advances in hybrid deep learning have improved the accuracy of short-term wind power forecasting, many state-of-the-art models usually consider signal denoising, residual decomposition, and attention mechanisms as independent modules without providing a unified solution. This paper proposes an end-to-end solution, W-HiTS-Attention (Wavelet Transform, N-stacked Hierarchical Interpolation for Time Series, Attention), which coherently integrates wavelet denoising, hierarchical residual learning from N-HiTS (Neural Hierarchical Interpolation), and an in-block self-attention mechanism. The proposed solution outperforms 21 benchmarks in accuracy, including state-of-the-art baselines such as N-BEATS, N-HiTS, TCN, Informer, Autoformer, LSTM, BiLSTM, GRU, and Prophet, achieving an RMSE of 55.56 W and an R2 of 0.9918. Furthermore, the results show that the proposed solution is efficient in terms of parameter count (0.033M), latency (0.0036 ms/sample), and training time, making it promising for low-latency inference in resource-constrained environments. The results show that the coherent integration of frequency preprocessing, hierarchical residual forecasting, and attention-based temporal refinement provides a robust, explainable, and deployable solution for short-term wind power forecasting.

Predictive control of Wind–Hydrogen coupled systems integrating secondary decomposition and PSO-AWOA

Accurate wind power forecasting is essential for the stability of modern power systems. However, current probabilistic forecasting frameworks often encounter a fundamental conflict between the computational efficiency required for high-dimensional meteorological pattern matching and the physical consistency of the resulting probability distributions. Existing methods frequently fail to maintain the logical monotonicity of quantiles or overlook the fine-grained temporal correlations in massive historical datasets. To address these critical gaps, this research develops a comprehensive framework that synergizes a hierarchical similarity filtering mechanism with a structurally constrained non-crossing quantile regression model. First, the target sample is partitioned into several weather segments, and a new two-stage high-similarity weather pattern matching method is developed to screen multiple sets of historical samples that are highly similar to the target weather pattern. Second, a deep learning model for probabilistic wind power quantile forecasting is proposed, which incorporates historical data augmentation. The model utilizes an attention mechanism to extract the correlation between the target and historical segments, while an improved non-crossing quantile regression model is adopted to ensure the validity of the output quantiles. Finally, the effectiveness of the proposed method is validated through case studies using real-world data from an actual wind farm.