ABSTRACT Accurate regional wind power forecasting (RWPF) is critical for grid integration and stability, yet remains challenging due to the inherently complex spatio‐temporal dependencies among geographically distributed wind farms. This paper introduces the spatio‐temporal dual‐encoder Transformer (ST‐DualFormer), an architecture designed to improve the accuracy of short‐term RWPF. ST‐DualFormer utilizes two parallel encoder streams to separately model temporal and spatial dependencies from meteorological and historical power data. In contrast to graph‐based models that rely on predefined spatial connections, ST‐DualFormer leverages the attention mechanism to capture comprehensive and dynamic correlations across all wind farms in a fully connected manner, facilitating flexible and comprehensive spatio‐temporal correlations. Evaluated on real‐world data from 28 wind farms, ST‐DualFormer achieves a normalized mean absolute error (nMAE) of 5.25% and a normalized root mean squared error (nRMSE) of 7.53% for three‐day‐ahead forecasting, outperforming the tested graph‐based and Transformer‐based baselines. Additional validation on two Weather2K‐R regional subsets further provides initial evidence that the dual‐stream architecture can transfer beyond the original study region.

Reliable prediction of near surface wind over complex terrain is limited by the mismatch between the spatial resolution of operational forecasts and terrain controlled local wind variability. Here, we develop KiloGen, a diffusion posterior sampling framework for kilometre scale wind forecast enhancement. KiloGen learns a high resolution vector wind prior from Weather Research and Forecasting (WRF) model simulations and constrains posterior sampling with 25 km forecasts from the European Centre for Medium Range Weather Forecasts (ECMWF) at inference time. This formulation avoids paired ECMWF and WRF training samples and an explicitly learned mapping from coarse to fine resolution. Applied over Shanxi, China, a region with complex mountainous terrain, KiloGen reconstructs terrain organized wind structures and restores high wavenumber variability while retaining the large scale evolution of the operational forecast. Station verification shows that KiloGen achieves the lowest overall wind speed root mean square error (RMSE) among the evaluated products, with larger benefits at elevated and topographically complex sites. The improvement is strongest under strong wind conditions, reducing RMSE by approximately 10

With the advancement of China’s “dual carbon” targets, Desert–Gobi–Wilderness (DGW) regions have become strategic areas for large-scale renewable energy deployment. However, the intermittency and variability of wind and solar resources pose challenges to power system stability, necessitating systematic evaluation of their characteristics and complementarity. This study uses ERA5 reanalysis data (2013–2023) to assess wind and solar resources in the Badain Jaran and Kumtag Deserts. A multi-dimensional framework is developed, incorporating availability, intermittency, variability, and complementarity, and a GIS-based multi-criteria decision-making method is applied for site selection. Results show that the Badain Jaran Desert is characterised by strong wind resources (average wind power density: 235.16 W/m2) and is suitable for wind-dominated development, whereas the Kumtag Desert exhibits superior solar resources (221.08 W/m2), favouring photovoltaic deployment. Significant wind–solar complementarity is identified, particularly in the central-western Badain Jaran and northeastern Kumtag regions. Three high-suitability sites were identified, including two in the Badain Jaran Desert and one in the Kumtag Desert, all characterised by favourable topographic conditions and high engineering feasibility. This study provides a scientific basis and a methodological framework for the planning of wind–solar hybrid systems and coordinated ecological development in DGW regions.

With the increasing integration of renewable energy sources, the power distribution terminals and data volumes in smart distribution networks continue to expand, which results in cloud architecture, fully distributed algorithms and some traditional communication methods failing to meet the operational requirements with the smart distribution power networks. In comparison, 5G communication exhibits significant potential performance due to its enhanced capabilities, particularly its network slicing feature, which facilitates the adoption of multi-access edge computing (MEC) by classifying terminal services and information types, while MEC could enhance computational capabilities at distribution terminals. This paper validates the feasibility of edge computing deployment in 5G networks based on the OMNET++ platform and proposes a hybrid architecture designed for smart distribution power networks, combining the respective advantages of both cloud and edge computing, whose performance further analyzed through comparative studies with other algorithms. The results demonstrated that the proposed cloud-edge collaborative architecture in this paper was effective and showed the promising performance for the smart distribution power networks.

Weather Foundation Models (WFMs) have recently attracted significant attention for their exceptional performance and inference efficiency in global-scale weather forecasting. However, their coarse spatial resolution and inherent biases constrain their utility for station-level forecasting, which is crucial for applications such as renewable energy management and aviation safety. To address these limitations, we propose the Adaptive Spatiotemporal Alignment Fusion Network (ASTAFN), a novel framework designed for accurate station-level weather forecasting through the synergistic integration of WFMs and station observations. ASTAFN incorporates two complementary data sources: (1) recent station observations, which offer fine-grained local trend information, and (2) WFM-generated forecasts, which provide broad-scale weather patterns. The core innovation of ASTAFN lies in its proxy station learning mechanism, which aligns the spatial structure and corrects the biases of WFMs relative to actual station data, facilitating the extraction of homogeneous spatiotemporal features from both sources. These features are dynamically fused at each forecasting step using an adaptive strategy, effectively compensating for WFM biases and enhancing predictive accuracy. Experimental evaluations on three real-world datasets demonstrate that ASTAFN reduces mean absolute error by 20%-35% compared to baseline WFMs for station-level wind speed forecasting. ASTAFN has been deployed on the regional station-level weather forecasting and analysis platform of the Chinese Academy of Meteorological Sciences, currently serving the Guangdong and Yunnan provinces in southern China.

Accurate forecasting of heatwaves is critical for ensuring the safe operation of electricity grids. Focusing on the complex terrain of Sichuan, China, this study investigates the optimization of spectral nudging parameters within the Weather Research and Forecasting (WRF) model to improve predictions of heatwave events. To overcome the subjectivity inherent in the traditional selection of the spectral nudging cutoff wavenumber, we propose an objective method based on power-spectrum energy diagnostics of the background field. This method determines an optimal domain-specific cutoff wavenumber. A series of sensitivity experiments were designed for a significant heatwave event that affected the Sichuan electricity grid in August 2019. These experiments evaluated the impact of different spectral nudging configurations, which considered varying domain sizes and forecast lead times, on correcting large-scale circulation drift and enhancing near-surface air temperature forecasts. The results demonstrate the following: (1) For a smaller domain or a longer forecast lead time, spectral nudging effectively compensates for circulation drift induced by weakening lateral boundary constraints, significantly improving the forecast of heatwave intensity and spatial extent, representing a compensatory effect. (2) For a larger domain that already adequately resolves large-scale circulation evolution, spectral nudging can over-constrain the model's internal dynamical processes, thereby degrading forecast performance, an outcome termed the over-constraint effect. (3) The proposed energy-threshold method provides an objective, physics-based strategy for identifying dominant large-scale waves and optimizing the spectral nudging cutoff wavenumber. This work offers practical insights for the operational application of spectral nudging over complex terrain to advance extreme temperature forecasting.

This study investigates the impacts of different physics schemes on hurricane forecasting across different applications of the unified forecast system (UFS) limited area model (LAM), including the UFS short-range weather (SRW) application, the hurricane analysis and forecast system (HAFS), and the UFS single-column model (SCM). The physics schemes compared here encompass cloud microphysics schemes, planetary boundary layer (PBL) parameterizations, surface layer schemes, gravity wave schemes, and land surface models. All these schemes are implemented through the common community physics package (CCPP) framework. Six experiments, each employing different combinations of physics schemes, were designed to explore their influences on hurricane forecasting. The simulated results from the UFS LAM applications are compared with the observed results from stations, radar, and satellite data. The impacts of various physics schemes on hurricane intensity, track, timing, location, and precipitation are investigated through case studies of different hurricanes, such as Barry, Lorenzo, and Ian. All experiments exhibit a right-of-track bias for hurricanes Barry and Lorenzo. In contrast, the forecasts for hurricane Ian have a left-of-track bias. The results show that the moist turbulence kinetic energy (TKE)-based eddy-diffusivity mass-flux (EDMF) PBL scheme can enhance the hurricane intensity through the moist process. The aerosol-aware (AA) Thompson microphysics scheme appears to provide better precipitation forecasts compared to those from the GFDL microphysics scheme. The combination of these updated physics schemes gives a better track comparing with the result from other old physics settings for hurricane Barry, and it also has an improved intensity in the forecast of hurricane Ian. Sensitivity tests conducted using the UFS SRW application indicate that increasing horizontal resolution and adding vertical levels can enhance model performance. Experiments also revealed that the PBL scheme significantly affects hurricane tracks, a finding further supported by SCM results. Additionally, sensitivity tests demonstrate that improved initial conditions through data assimilation can lead to better simulations of both the hurricane track and intensity, as well as more accurate predictions of hurricane landfall timing and location.

Aiming at the challenges of complex spatial-temporal correlation and strong nonlinearity in the power prediction of large-scale wind farm clusters, this study proposes a short-term wind power prediction method that combines a dynamic graph structure and a Kolmogorov-Arnold Network (KAN) enhanced neural network. Firstly, a spectral embedding fuzzy C-means (FCM) cluster partition method combining geographic location and numerical weather prediction (NWP) is proposed to solve the problem of insufficient spatio-temporal representation ability of traditional methods. Secondly, a dynamic directed graph construction mechanism based on a stacked wind direction matrix and wind speed mutual information is designed to describe the directional correlation between stations with the evolution of meteorological conditions. Finally, a prediction model of dynamic graph convolution and Transformer based on KAN enhancement (DGTK-Net) is constructed to improve the fitting ability of complex nonlinear relationships. Based on the cluster data of 31 wind farms in Gansu Province of China and the cluster data of 70 wind farms in Inner Mongolia, a case study is carried out. The results show that the proposed model is significantly better than the comparison methods in terms of key evaluation indicators, and the root mean square error is reduced by about 1.16% on average. This method provides a prediction tool that can adapt to time and space changes for engineering practice, which is helpful to improve the wind power consumption capacity and operation economy of the power grid.

Artificial intelligence vision models have been widely applied in electric power scenarios such as transmission line inspection and operational safety management. However, existing models still face challenges such as limited accuracy and weak model generalization ability. This study aims to address these issues by constructing and pre-training a backbone for deep vision models in electric power scenarios. We collected a large number of images from the electric power domain and combined them with general-domain images for unsupervised pre-training, thereby improving the downstream task accuracy of the model. We propose several improvements to enhance the performance and training efficiency for visual backbones in power scenarios, including a multi-scale visual backbone combining convolutional and Transformer blocks, a rotational position embedding method for 2D images, a local self-attention operator based on the Flash Attention algorithm, an unsupervised pre-training algorithm which learns to reconstruct the features from a contrastive language-image (CLIP) model, and a grouped masking method for masked image modeling pre-training with both convolutional and Transformer architectures. We evaluate the performance of the proposed backbone model across multiple downstream scenarios and datasets.

Accurate multistep ultrashort-term wind power forecasting for wind farm clusters is essential for secure and flexible power system operation. However, existing forecasting methods often rely on pairwise spatial correlations or static graph structures, which limits their ability to represent multiscale spatial dependencies and transient fluctuation patterns among wind farms. In addition, the temporal mismatch between future numerical weather prediction (NWP) information and different forecasting horizons may further degrade multistep forecasting accuracy. To address these challenges, this study proposes a hierarchical dynamic hypergraph forecasting framework with input-aware meteorological modeling for wind farm cluster power forecasting. Specifically, four types of hyperedges are constructed to describe local correlations, regional coordinated structures, global similarity patterns, and transient response relationships among wind farms. A multichannel architecture is then designed to jointly extract historical power information, future NWP features, graph-based pairwise spatial representations, and hypergraph-based high-order spatial representations. Furthermore, an input-aware modeling strategy is developed to adaptively incorporate horizon-related meteorological information, thereby reducing temporal mismatch in multistep forecasting. The proposed method is validated using real-world data from large-scale wind farm clusters in China. Experimental results averaged over multiple independent trials show that the proposed method achieves lower NRMSE and NMAE than representative transformer-based and graph-based benchmark models, with an average absolute NRMSE reduction of 1.5% over the benchmark models and a 13.1% relative NRMSE reduction compared with the strongest baseline under the tested setting. The results indicate that the proposed framework can improve multistep forecasting accuracy and curve-tracking consistency.

Accurate station-scale weather forecasts support renewable-energy and transportation operations, yet gridded numerical weather prediction (NWP) is affected by representativeness errors and regional biases when transferred to irregularly distributed stations. We propose TF-STNet, a time–frequency dual-branch spatiotemporal network for NWP-to-station bias correction. A station-centered K-nearest-neighbor (KNN) operator retains multiple local NWP trajectories. The time-domain pathway separately encodes recent observations and future NWP, aligns them over the forecast horizon using gated dilated causal convolutions, and propagates lead-resolved states through a coordinate-conditioned directed station graph. The frequency-domain pathway learns spectral weights and applies separate attention to amplitude and phase across neighboring NWP cells. Prediction-level fusion combines the two station forecasts by variable, station, and lead time. The evaluation uses hourly data for wind speed, pressure, relative humidity, and temperature from 455 stations in Hebei, Shandong, Fujian, and Sichuan. Across five independent runs on 16 region–variable tasks, TF-STNet achieves the lowest mean absolute error (MAE) in 15 tasks and the highest Pearson correlation coefficient (PCC) in 15 tasks; its pressure MAE reduction relative to the strongest learned comparator ranges from 16.2% to 57.4% across the four regions. It has lower MAE than raw NWP in seven of eight high-wind or rapid-change event tests and than simple pressure model-output-statistics corrections in all four regions. The Shandong–temperature task and the Hebei–high-wind case illustrate the limits of the present point-forecast formulation.

We proposed a novel historical feature reuse scheme to improve the accuracy of photovoltaic power forecasting model. Firstly, a weather type classification method based on the Elkan K-means algorithm was proposed, and a feature matching mechanism based on Markov distance was constructed to fuse information; Then, a bidirectional recurrent residual network was constructed, which improved the feature extraction performance of the forecasting model for different photovoltaic output scenarios; Finally, an error decoupling mechanism was proposed to evaluate the upper limit of the forecasting accuracy of the model. Taking the data provided by a photovoltaic power station in Jilin Province, China as the research object, the day-ahead power forecasting accuracy is 91.12 %, verifying the validity of the proposed model.

Firm power generation refers to renewable energy that can provide dispatchable power comparable to conventional coal-fired generation, satisfying the load profile on a 24/365 basis. Achieving firm power generation relies on flexibility resource configurations that integrate generation diversity, energy storage, and other strategies to compensate for the inherent variability of renewable energy. Although firm power generation represents the ultimate form of a zero-carbon power system, the transition pathway from coal-fired generation to firm power generation remains underexplored because coal-fired generation continues to serve as the backbone of power system operation. In this regard, this work proposes a multi-stage optimization framework that combines coal plant retirement with firm renewable deployment, enabling a systematic decarbonization pathway for the power system. The optimal results illustrate that the system levelized cost of electricity increases from 48.26 $/MWh in the coal-dominated system to 94.01 $/MWh in the firm renewable system to meet load demand on a 24/365 basis. Additionally, a firm kWh premium is introduced to quantify the additional cost of delivering firm power generation from renewable energy. The results show that the premium reaches 2.50 in the firm renewable system, while the combined deployment of multiple storage (e.g., battery, pumped hydro, and hydrogen) reduces it by 13.20%. The proposed framework provides an evolutionary pathway for coordinating coal power phaseout with firm renewable expansion, offering valuable insights for cost-effective power system decarbonization.

The spatial and temporal loss of information caused by the conversion of polar to cartesian weather radar data has been quantised for Chenies radar which forms part of the United Kingdom weather radar network. This analysis was carried out in the frequency domain by the use of the Discrete Fourier Transform (DFT). sido quantizada para el radar Chenies el cual forma parte de la red de radares del Reino Unido. Este analisis fue llevado a cabo en el dominio de la frecuencia usando la transformada discreta de Fourier (DFT).

This two-year trial aims to bring together academics and industrial partners from UK and China to conduct a pilot study on the use of the active phased array radar to provide early urban flood warnings for Chinese mega cities, which facing challenging urban flood issues. This is the first in the world of cascade modelling using the cutting-edge active phase array radar (APRA) to provide rainfall monitoring and nowcasting information for a real-time two-dimension urban drainage model. The collaboration built up by this project and the first-hand experiment data will serve well to further catalyse the taking-up of state-of-the-art weather radars for urban flood risk management, and to tackle the innovation in tuning the radar technology to fit the complex urban environment as well as advanced modelling facilities that are designed to link the observations, providing decision making support to the city government. Recommendations for applying high spatial-temporal resolution precipitation data to real-time flood forecasting on an urban catchment are provided and suggestions for further investigation are discussed.

Recognizing the intricate spatiotemporal correlation (STC) among wind farms (WFs) is critical to achieving better predictions for wind farm clusters (WFCs). To describe the STC accurately, this paper employs the wind angular field method to transform the wind series of WFs into different 2-D feature maps, and then construct homogeneous and heterogeneous STC graphs from these maps. The graphs are dynamically updated at the frequency of data update to capture time-varying STC among WFs. Finally, a dynamic graph attention network, designed according to the STC graphs, is established for WFC prediction. Through the above process, dynamic and accurate descriptions of STC are realized in WFC prediction. From the case study of a large-scale WFC with a capacity over 5800 MW in Northeast China, the proposed method reduces the root mean square error of the prediction in the next 24 hours by 2.67%.

Accurate wind power forecasting (WPF) is essential for ensuring the secure operation of power systems and supporting the sustainable development of renewable energy. Against the backdrop of rapid renewable energy development, achieving high-accuracy forecasting under complex meteorological conditions and diverse application scenarios remains a central objective in this field. In recent years, deep representation learning models have become a research focus in WPF owing to their strengths in nonlinear modeling and the representation of complex data structures. This review provides a systematic overview of current progress in deep representation models, with particular emphasis on pre-trained large models, graph neural networks, and recurrent neural networks, and further summarizes the roles of attention mechanisms across different dimensions. Typical issues and solutions in data preprocessing are outlined, and various signal decomposition methods are comparatively analyzed. In addition, publicly available datasets are presented, experimental details of representative methods are synthesized, and challenges across diverse application scenarios are addressed. Building on current trends, this work proposes future research directions, aiming to provide a structured reference for subsequent studies in WPF.

Global wind power capacity, especially in China, is booming, with new farms spanning diverse terrains and climates. The industry urgently needs accurate wind power foundation models to shorten commissioning and accelerate grid connection. This is because site-specific time series models (TSMs) are not well suited to data-scarce scenarios and generalize poorly, while generic large time series models (LTSMs) are mostly limited to univariate inputs and cannot fully exploit static site attributes or the dependencies between power and meteorological covariates, leading to insufficient accuracy. To fill this gap, we propose Tyan-WP, the first wind power foundation model for ultra-short-term probabilistic forecasting. Pretrained on a large-scale wind power dataset covering more than 126,000 U.S. sites over seven years, Tyan-WP further improves zero-shot forecasting through two domain-specific module designs: static site embedding using coordinate, terrain, and ecoregion metadata, and a power-aware meteorological fusion (PAMF) module that models interactions between historical power and meteorological covariates. Under a unified evaluation protocol, Tyan-WP surpasses eight site-specific supervised TSMs on 10 in-domain sites and outperforms eleven generic LTSMs on 127 in-domain sites, reducing MAE by 19.9

Driven by the dual-carbon goals, photovoltaic (PV) battery systems at renewable energy stations are increasingly clustered on the distribution side. The rapid expansion of these clusters, together with the pronounced uncertainty and spatio-temporal heterogeneity of PV generation, degrades battery utilization and forces conservative dispatch. To address this, we propose a “spatio-temporal clustering–deep estimation” framework for short-term interval forecasting of PV clusters. First, a graph is built from meteorological–geographical similarity and partitioned into sub-clusters by a self-supervised DAEGC. Second, an attention-based spatio-temporal graph convolutional network (ASTGCN) is trained independently for each sub-cluster to capture local dynamics; the individual forecasts are then aggregated to yield the cluster-wide point prediction. Finally, kernel density estimation (KDE) non-parametrically models the residuals, producing probabilistic power intervals for the entire cluster. At the 90% confidence level, the proposed framework improves PICP by 4.01% and reduces PINAW by 7.20% compared with the ASTGCN-KDE baseline without spatio-temporal clustering, demonstrating enhanced interval forecasting performance.

Accurate forecasting of meteorological variables is crucial for reliable wind power prediction, ensuring grid security and efficient energy dispatch amid growing renewable energy deployment. Traditional physics-based models are limited by incomplete mechanisms and high computational costs, while purely data-driven approaches struggle to capture the physical constraints inherent in atmospheric evolution. As a result, the accuracy of meteorological variable prediction is constrained, and robustness under extreme conditions, such as sudden wind changes, remains insufficient. To address this challenge, this paper proposes a hybrid forecasting framework that integrates physical mechanisms based on partial differential equations with data-driven approaches. The framework incorporates weighted fusion and MLP-based fusion strategies at the input and output stages to enable collaborative optimization between physical priors and data-driven models, improving overall predictive performance and generalization ability. Experiments on the ERA5 reanalysis dataset and real wind farm observation data, covering single-step and multi-step forecasting tasks, demonstrate that the proposed method significantly enhances the accuracy and robustness of key meteorological variable predictions, including wind speed, wind direction, air pressure, and temperature. In regions with high and highly fluctuating wind speeds, the MAE of prediction improves by 30.4 percent and 33.7 percent. These predictions provide a reliable foundation for wind power forecasting, and thus contribute to increased power prediction accuracy.