Similar papers
Conference
Aug 2026
Short-Term Wind Power Forecasting Using Transformer-XGBoost Hybrid Model
Accurate wind forecasting is critical to ensure stable and efficient integration of renewable energy resources in modern power systems. However, the inherent variability and non-stationarity of wind pose a significant forecasting problem for modern power system operators to ensure power system stability. A new hybrid forecasting model incorporating a Transformer Encoder and XGBoost regressor is proposed in this paper to enhance one-hour-ahead short-term forecasting of wind power generation. The model employs Transformer Encoders to exploit complex temporal relationships from sequences of 24-hour past wind data, learning daily and periodic trends, ramps, and hidden dynamics. The generated temporal feature vectors are then combined with meteorological variables such as wind speed and direction at various altitudes and passed on to the XGBoost model to obtain regression forecasts. The proposed approach has been validated on real-world datasets obtained from commercial wind energy farms. Experimental results demonstrate that the hybrid model outperforms standalone Transformer and XGBoost models by achieving accurate values for mean squared error, mean absolute error and coefficient of determination. In addition, model has tested for long-term forecasting as well. Although the model is effective for short-term forecasting, due to increased uncertainty and error propagation, its performance degrades with extended forecasting horizons.
Heshan Senapriya, Sakun Rasilka, D. P. Wadduwage
· Moratuwa Engineering Researc... · 0 citations
Open access
Sep 2026
Initial-State-Aware Multi-Scale Transformer for Short-Term Wind Power Forecasting
Accurate short-term wind power forecasting is essential for the secure and economic operation of power systems with high renewable energy penetration. However, forecasting performance is still affected by meteorological forecast uncertainty and the complex multi-scale fluctuation characteristics of wind power generation. To address these challenges, this paper proposes an Initial-State-Aware Multi-Scale Transformer framework for 12 h-ahead wind power forecasting. The principal methodological contribution is an initial-state-aware meteorological representation and progressive fusion strategy tailored to weather-driven wind power forecasting. The framework explicitly distinguishes the atmospheric state available at forecast initialization from the subsequent forecast meteorological trajectory and uses the former to condition the representation of the latter through cross-attention and gated residual fusion. The resulting meteorological representation is then progressively coupled with coarse- and fine-scale historical power representations, and a horizon-oriented forecasting head generates the future power sequence in parallel. Experiments on three wind farms demonstrate that the proposed method achieves the best overall forecasting performance. Compared with the strongest baseline model, it reduces NMAE and NRMSE by 8.54% and 2.36%, respectively.
Chao-Ying Yang, Jun Zhao, Peng Han et al.
· Energies · 0 citations
Open access
Sep 2026
Empowering Wind Energy Output Optimization: Comparative Assessment of Hybrid Artificial Intelligence Models Towards Wind Speed Forecasting Accuracy
Wind power forecasting is essential for the reliable and efficient operation of wind farms based on power grids, which is significantly important for stakeholders like wind farm owners, power pools, and power traders. Wind power prediction is also crucial for optimal power dispatch, grid security, and minimizing generation curtailment at wind farms. Hence, an accurate methodology for the power production of wind farms is needed for identifying long-term operational performance, failure detection, and ensuring grid integration. The present study proposes a hybrid machine learning (ML) approach that combines the strengths of random forest regression (RFR), artificial neural network (ANN), and support vector regression (SVR), using linear regression (LR) as a meta-model for wind speed forecasting. This modeling approach was trained, validated, and tested on 87,600 hourly data points across 10 variables from high-wind potential sites in the Kingdom of Saudi Arabia: Damat Al Jandal, Taif, Abha, East Coast, and Red Sea. The key findings suggest that the hybrid RFR + ANN + SVR model performed better than individual models, including the simple persistence model, achieving R2 scores up to 95.3 in testing, with train–test performance loss of less than 5% across all scenarios. The mean bias error (MBE) stayed within ±0.17 m/s, and the relative root mean square error (RRMSE) stayed below 15%. The hybrid model outperformed the metric-site standalone models and persistence model. Through feature importance analysis, this study also finds that temperature, relative humidity, and cyclical time-encoded features are the most important inputs. For offshore sites, thermal features like pressure and dew point were found to be more important. Due to its demonstrated superior performance across metric-site combinations, this hybrid framework is adaptable to other wind regions with intermittent renewables, with implications for reliable renewable energy integration and grid load stability.
Latif Bukari Rashid, S. Rehman, Mohamad E. Zayed et al.
· Energies · 0 citations
Open access
Sep 2026
Does Machine Learning Improve Wind Power Forecasting? An Experimental Investigation
Accurate wind power forecasting is essential for the stable and economic operation of power systems with high renewable penetration. Although machine learning models have been widely adopted for this task, the assumption that greater model complexity invariably yields superior forecasting accuracy has received insufficient scrutiny. This paper presents a systematic experimental investigation that covers two complementary stages, i.e., wind speed correction and wind power forecasting. For wind speed correction, we compare 10 machine learning methods, including spanning linear, instance-based, and tree-based ensemble learners, under four newly proposed progressively enriched feature configurations. For wind power forecasting, we benchmark 20 methods spanning traditional machine learning, time-series deep learning, and Transformer-based architectures on two geographically distinct wind farms. Our results reveal a clear task-dependent pattern. In wind speed correction, tree-based ensemble methods, particularly gradient boosting variants, consistently dominate, and feature engineering contributes more to accuracy gains than model selection. In wind power forecasting, deep learning architectures substantially and consistently outperform traditional methods, with attention-based models generalizing the most robustly across regimes and recurrent networks proving to be the most sensitive to regime shifts. These findings provide actionable task-specific guidance for model selection in operational wind power forecasting systems.
Zhi-Min Li, Yu Chen, Tingzhao Yu et al.
· Forecasting · 0 citations
Open access
Aug 2026
Transformer-based short-term forecasting of renewable power and grid load using high-resolution weather data with explainable AI
Accurate forecasting of electricity load, solar power, and wind power is critical to the reliability of modern power systems, particularly in regions such as California with high penetration of variable power sources. California’s power system is managed by the California Independent System Operator (CAISO), which operates a grid that receives a significant portion of its power from various renewable energy sources. A novel forecasting method based on the Transformer architecture is introduced to accurately forecast electricity load, solar power, and wind power. This study will utilize an interval data set that is collected every 10 min. A new forecasting methodology has been developed by combining CAISO’s real-time operational data with high-frequency meteorological datasets from the National Renewable Energy Laboratory (NREL) National Solar Radiation Database (NSRDB). Since the forecasting model was tested using historical data, we used a chronological data split. We also engineered all feature combinations using historical data. Three different transformers were developed specifically for each task: Load, Solar Power, and Wind Power. Each transformer includes positional encoding, multi-head self-attention and temporal characteristics. Testing demonstrated strong performance of the forecasting models, as evidenced by average R2 values of approximately 0.95 for Load, 0.80 for Solar, and 0.77 for Wind. The results indicate the inherent volatility of renewable energy generation and confirm that an attention-based architecture can be utilised to represent complex time-series relationships. The proposed solution offers a viable approach to developing a scalable and reliable short-term energy-forecasting platform for renewable-integrated power systems.
Dajiang Hu, Na Liu
· PeerJ Computer Science · 0 citations
Open access
Aug 2026
An MT-Transformer Framework for Coordinated Wind-Solar-Load Forecasting with Net-Load-Based Coal-Power Regulation Demand Identification
Data-driven forecasting has become increasingly important for describing the temporal interactions among heterogeneous variables in modern power systems. To capture the nonlinear coupling, heterogeneous fluctuations, and multi-scale temporal variations between renewable generation and load demand, this study develops an MT-Transformer framework for coordinated wind–solar–load forecasting. Meteorological variables, historical renewable output, historical load, and temporal labels are integrated as model inputs. A shared Transformer encoder is used to learn common temporal representations, and task-specific forecasting heads are designed to generate synchronized predictions for wind power, PV power, and load. The experimental results show that MT-Transformer achieves an MAE of 0.0848, an RMSE of 0.1254, and an R² of 0.9302. Compared with the Persistence Model, the MAE and RMSE decrease by 36.05% and 30.49%, respectively. The predicted outputs are further converted into a net-load sequence, from which fluctuation indicators are derived. The peak–valley difference reaches 0.462 p.u., and the maximum ramp rate reaches 0.087 p.u./h, indicating evident peak-shaving pressure and short-term regulation demand. These findings confirm that the proposed framework improves coordinated forecasting performance and provides quantitative evidence for coal-power peak regulation, reserve capacity allocation, and ancillary service demand identification.
Meng Huang, Lei Wang, Teng Luo et al.
· EAI Endorsed Transactions on... · 0 citations