It is fundamental and useful to investigate how deep learning forecasting models(DLMs)perform compared to operational oceanography forecast systems(OFSs).However,few studies have intercompared their performances using...It is fundamental and useful to investigate how deep learning forecasting models(DLMs)perform compared to operational oceanography forecast systems(OFSs).However,few studies have intercompared their performances using an identical reference.In this study,three physically reasonable DLMs are implemented for the forecasting of the sea surface temperature(SST),sea level anomaly(SLA),and sea surface velocity in the South China Sea.The DLMs are validated against both the testing dataset and the“OceanPredict”Class 4 dataset.Results show that the DLMs'RMSEs against the latter increase by 44%,245%,302%,and 109%for SST,SLA,current speed,and direction,respectively,compared to those against the former.Therefore,different references have significant influences on the validation,and it is necessary to use an identical and independent reference to intercompare the DLMs and OFSs.Against the Class 4 dataset,the DLMs present significantly better performance for SLA than the OFSs,and slightly better performances for other variables.The error patterns of the DLMs and OFSs show a high degree of similarity,which is reasonable from the viewpoint of predictability,facilitating further applications of the DLMs.For extreme events,the DLMs and OFSs both present large but similar forecast errors for SLA and current speed,while the DLMs are likely to give larger errors for SST and current direction.This study provides an evaluation of the forecast skills of commonly used DLMs and provides an example to objectively intercompare different DLMs.展开更多
Load forecasting is of great significance to the development of new power systems.With the advancement of smart grids,the integration and distribution of distributed renewable energy sources and power electronics devi...Load forecasting is of great significance to the development of new power systems.With the advancement of smart grids,the integration and distribution of distributed renewable energy sources and power electronics devices have made power load data increasingly complex and volatile.This places higher demands on the prediction and analysis of power loads.In order to improve the prediction accuracy of short-term power load,a CNN-BiLSTMTPA short-term power prediction model based on the Improved Whale Optimization Algorithm(IWOA)with mixed strategies was proposed.Firstly,the model combined the Convolutional Neural Network(CNN)with the Bidirectional Long Short-Term Memory Network(BiLSTM)to fully extract the spatio-temporal characteristics of the load data itself.Then,the Temporal Pattern Attention(TPA)mechanism was introduced into the CNN-BiLSTM model to automatically assign corresponding weights to the hidden states of the BiLSTM.This allowed the model to differentiate the importance of load sequences at different time intervals.At the same time,in order to solve the problem of the difficulties of selecting the parameters of the temporal model,and the poor global search ability of the whale algorithm,which is easy to fall into the local optimization,the whale algorithm(IWOA)was optimized by using the hybrid strategy of Tent chaos mapping and Levy flight strategy,so as to better search the parameters of the model.In this experiment,the real load data of a region in Zhejiang was taken as an example to analyze,and the prediction accuracy(R2)of the proposed method reached 98.83%.Compared with the prediction models such as BP,WOA-CNN-BiLSTM,SSA-CNN-BiLSTM,CNN-BiGRU-Attention,etc.,the experimental results showed that the model proposed in this study has a higher prediction accuracy.展开更多
Weather forecasts from numerical weather prediction models play a central role in solar energy forecasting,where a cascade of physics-based models is used in a model chain approach to convert forecasts of solar irradi...Weather forecasts from numerical weather prediction models play a central role in solar energy forecasting,where a cascade of physics-based models is used in a model chain approach to convert forecasts of solar irradiance to solar power production.Ensemble simulations from such weather models aim to quantify uncertainty in the future development of the weather,and can be used to propagate this uncertainty through the model chain to generate probabilistic solar energy predictions.However,ensemble prediction systems are known to exhibit systematic errors,and thus require post-processing to obtain accurate and reliable probabilistic forecasts.The overarching aim of our study is to systematically evaluate different strategies to apply post-processing in model chain approaches with a specific focus on solar energy:not applying any post-processing at all;post-processing only the irradiance predictions before the conversion;post-processing only the solar power predictions obtained from the model chain;or applying post-processing in both steps.In a case study based on a benchmark dataset for the Jacumba solar plant in the U.S.,we develop statistical and machine learning methods for postprocessing ensemble predictions of global horizontal irradiance(GHI)and solar power generation.Further,we propose a neural-network-based model for direct solar power forecasting that bypasses the model chain.Our results indicate that postprocessing substantially improves the solar power generation forecasts,in particular when post-processing is applied to the power predictions.The machine learning methods for post-processing slightly outperform the statistical methods,and the direct forecasting approach performs comparably to the post-processing strategies.展开更多
The China Meteorological Administration(CMA)said that in the last five years,China has made big improvements in its weather services.This includes better weather forecasts and ways to protect people from disasters.
The dynamics of vapor−liquid−solid(V−L−S)flow boiling in fluidized bed evaporators exhibit inherent complexity and chaotic behavior,hindering accurate prediction of pressure drop signals.To address this challenge,this...The dynamics of vapor−liquid−solid(V−L−S)flow boiling in fluidized bed evaporators exhibit inherent complexity and chaotic behavior,hindering accurate prediction of pressure drop signals.To address this challenge,this study proposes an innovative hybrid approach that integrates wavelet neural network(WNN)with chaos analysis.By leveraging the Cross-Correlation(C−C)method,the minimum embedding dimension for phase space reconstruction is systematically calculated and then adopted as the input node configuration for the WNN.Simulation results demonstrate the remarkable effectiveness of this integrated method in predicting pressure drop signals,advancing our understanding of the intricate dynamic phenomena occurring with V−L−S fluidized bed evaporators.Moreover,this study offers a novel perspective on applying advanced data-driven techniques to handle the complexities of multi-phase flow systems and highlights the potential for improved operational prediction and control in industrial settings.展开更多
Time series forecasting is important in the fields of finance,energy,and meteorology,but traditional methods often fail to cope with the complex nonlinear and nonstationary processes of real data.In this paper,we prop...Time series forecasting is important in the fields of finance,energy,and meteorology,but traditional methods often fail to cope with the complex nonlinear and nonstationary processes of real data.In this paper,we propose the FractalNet-LSTM model,which combines fractal convolutional units with recurrent long short-term memory(LSTM)layers to model time series efficiently.To test the effectiveness of the model,data with complex structures and patterns,in particular,with seasonal and cyclical effects,were used.To better demonstrate the obtained results and the formed conclusions,the model performance was shown on the datasets of electricity consumption,sunspot activity,and Spotify stock price.The result showed that the proposed model outperforms traditional approaches at medium forecasting horizons and demonstrates high accuracy for data with long-term and cyclical dependencies.However,for financial data with high volatility,the model’s efficiency decreases at long forecasting horizons,indicating the need for further adaptation.The findings suggest further adaptation.The findings suggest that integrating fractal properties into neural network architecture improves the accuracy of time series forecasting and can be useful for developing more accurate and reliable forecasting systems in various industries.展开更多
Forecasting tropical cyclone(TC)activities has been a topic of great interest and research.Taiwan Island(TW)is one of the key regions that is highly exposed to TCs originated from the western North Pacific.Here,the au...Forecasting tropical cyclone(TC)activities has been a topic of great interest and research.Taiwan Island(TW)is one of the key regions that is highly exposed to TCs originated from the western North Pacific.Here,the authors utilize two mainstream reanalysis datasets for the period 1979-2013 and propose an effective statistical seasonal forecasting model-namely,the Sun Yat-sen University(SYSU)Model-for predicting the number of TC landfalls on TW based on the environmental factors in the preseason.The comprehensive predictor sampling and multiple linear regression show that the 850-hPa meridional wind over the west of the Antarctic Peninsula in January,the 300-hPa specific humidity over the open ocean southwest of Australia in January,the 300-hPa relative vorticity over the west of the Sea of Okhotsk in March,and the sea surface temperature in the South Indian Ocean in April,are the most significant predictors.The correlation coefficient between the modeled results and observations reaches 0.87.The model is validated by the leave-one-out and nine-fold cross-validation methods,and recent 9-yr observations(2014-2022).The Antarctic Oscillation,variabilities of the western Pacific subtropical high,Asian summer monsoon,and oceanic tunnel are the possible physical linkages or mechanisms behind the model result.The SYSU Model exhibits a 98%hit rate in 1979-2022(43 out of 44),suggesting an operational potential in the seasonal forecasting of TC landfalls on TW.展开更多
Based on ground observation data of relative humidity,the prediction performance of STNF and MIFS in each competition area during February 13-26,2024 was tested and evaluated by using two intelligent forecasting metho...Based on ground observation data of relative humidity,the prediction performance of STNF and MIFS in each competition area during February 13-26,2024 was tested and evaluated by using two intelligent forecasting methods(STNF and MIFS).The results show that STNF had better performance in forecasting relative humidity in high-altitude areas,and was suitable for fine forecasting under complex terrain.MIFS improved the short-term forecast of some low-altitude stations,but the long-term reliability was insufficient.STNF method performed better than MIFS during 0-24 h.As the prediction time extended to 24-72 h,the errors of both methods showed a systematic increase trend.STNF had higher precision,lower root mean square error and smaller mean error in most regions under the background of most weather systems,showing its superiority as a forecasting method of relative humidity.However,the precision of MIFS was slightly higher than that of STNF in Liangcheng without system background,revealing that MIFS may also be an effective option in some specific conditions.展开更多
Accurate Electric Load Forecasting(ELF)is crucial for optimizing production capacity,improving operational efficiency,and managing energy resources effectively.Moreover,precise ELF contributes to a smaller environment...Accurate Electric Load Forecasting(ELF)is crucial for optimizing production capacity,improving operational efficiency,and managing energy resources effectively.Moreover,precise ELF contributes to a smaller environmental footprint by reducing the risks of disruption,downtime,and waste.However,with increasingly complex energy consumption patterns driven by renewable energy integration and changing consumer behaviors,no single approach has emerged as universally effective.In response,this research presents a hybrid modeling framework that combines the strengths of Random Forest(RF)and Autoregressive Integrated Moving Average(ARIMA)models,enhanced with advanced feature selection—Minimum Redundancy Maximum Relevancy and Maximum Synergy(MRMRMS)method—to produce a sparse model.Additionally,the residual patterns are analyzed to enhance forecast accuracy.High-resolution weather data from Weather Underground and historical energy consumption data from PJM for Duke Energy Ohio and Kentucky(DEO&K)are used in this application.This methodology,termed SP-RF-ARIMA,is evaluated against existing approaches;it demonstrates more than 40%reduction in mean absolute error and root mean square error compared to the second-best method.展开更多
In this paper,we introduce TianXing,a transformer-based data-driven model designed with physical augmentation for skillful and efficient global weather forecasting.Previous data-driven transformer models such as Pangu...In this paper,we introduce TianXing,a transformer-based data-driven model designed with physical augmentation for skillful and efficient global weather forecasting.Previous data-driven transformer models such as Pangu-Weather,FengWu,and FuXi have emerged as promising alternatives for numerical weather prediction in weather forecasting.However,these models have been characterized by their substantial computational resource consumption during training and limited incorporation of explicit physical guidance in their modeling frameworks.In contrast,TianXing applies a linear complexity mechanism that ensures proportional scalability with input data size while significantly diminishing GPU resource demands,with only a marginal compromise in accuracy.Furthermore,TianXing proposes an explicit attention decay mechanism in the linear attention derived from physical insights to enhance its forecasting skill.The mechanism can reweight attention based on Earth's spherical distances and learned sparse multivariate coupling relationships,promptingTianXing to prioritize dynamically relevant neighboring features.Finally,to enhance its performance in mediumrange forecasting,TianXing employs a stacked autoregressive forecast algorithm.Validation of the model's architecture is conducted using ERA5 reanalysis data at a 5.625°latitude-longitude resolution,while a high-resolution dataset at 0.25°is utilized for training the actual forecasting model.Notably,the TianXing exhibits excellent performance,particularly in the Z500(geopotential height)and T850(temperature)fields,surpassing previous data-driven models and operational fullresolution models such as NCEP GFS and ECMWF IFS,as evidenced by latitude-weighted RMSE and ACC metrics.Moreover,the TianXing has demonstrated remarkable capabilities in predicting extreme weather events,such as typhoons.展开更多
Deep learning(DL)has revolutionized time series forecasting(TSF),surpassing traditional statistical methods(e.g.,ARIMA)and machine learning techniques in modeling complex nonlinear dynamics and long-term dependencies ...Deep learning(DL)has revolutionized time series forecasting(TSF),surpassing traditional statistical methods(e.g.,ARIMA)and machine learning techniques in modeling complex nonlinear dynamics and long-term dependencies prevalent in real-world temporal data.This comprehensive survey reviews state-of-the-art DL architectures forTSF,focusing on four core paradigms:(1)ConvolutionalNeuralNetworks(CNNs),adept at extracting localized temporal features;(2)Recurrent Neural Networks(RNNs)and their advanced variants(LSTM,GRU),designed for sequential dependency modeling;(3)Graph Neural Networks(GNNs),specialized for forecasting structured relational data with spatial-temporal dependencies;and(4)Transformer-based models,leveraging self-attention mechanisms to capture global temporal patterns efficiently.We provide a rigorous analysis of the theoretical underpinnings,recent algorithmic advancements(e.g.,TCNs,attention mechanisms,hybrid architectures),and practical applications of each framework,supported by extensive benchmark datasets(e.g.,ETT,traffic flow,financial indicators)and standardized evaluation metrics(MAE,MSE,RMSE).Critical challenges,including handling irregular sampling intervals,integrating domain knowledge for robustness,and managing computational complexity,are thoroughly discussed.Emerging research directions highlighted include diffusion models for uncertainty quantification,hybrid pipelines combining classical statistical and DL techniques for enhanced interpretability,quantile regression with Transformers for riskaware forecasting,and optimizations for real-time deployment.This work serves as an essential reference,consolidating methodological innovations,empirical resources,and future trends to bridge the gap between theoretical research and practical implementation needs for researchers and practitioners in the field.展开更多
Weather forecasting is crucial for agriculture,transportation,and industry.Deep Learning(DL)has greatly improved the prediction accuracy.Among them,Graph Neural Networks(GNNs)excel at processing weather data by establ...Weather forecasting is crucial for agriculture,transportation,and industry.Deep Learning(DL)has greatly improved the prediction accuracy.Among them,Graph Neural Networks(GNNs)excel at processing weather data by establishing connections between regions.This allows them to understand complex patterns that traditional methods might miss.As a result,achieving more accurate predictions becomes possible.The paper reviews the role of GNNs in short-to medium-range weather forecasting.The methods are classified into three categories based on dataset differences.The paper also further identifies five promising research frontiers.These areas aim to boost forecasting precision and enhance computational efficiency.They offer valuable insights for future weather forecasting systems.展开更多
Global climate change,along with the rapid increase of the population,has put significant pressure on water security.A water reservoir is an effective solution for adjusting and ensuring water supply.In particular,the...Global climate change,along with the rapid increase of the population,has put significant pressure on water security.A water reservoir is an effective solution for adjusting and ensuring water supply.In particular,the reservoir water level is an essential physical indicator for the reservoirs.Forecasting the reservoir water level effectively assists the managers in making decisions and plans related to reservoir management policies.In recent years,deep learning models have been widely applied to solve forecasting problems.In this study,we propose a novel hybrid deep learning model namely the YOLOv9_ConvLSTM that integrates YOLOv9,ConvLSTM,and linear interpolation to predict reservoir water levels.It utilizes data from Sentinel-2 satellite images,generated from visible spectrum bands(Red-Blue-Green)to reconstruct true-color reservoir images.Adam is used as the optimization algorithm with the loss function being MSE(Mean Squared Error)to evaluate the model’s error during training.We implemented and validated the proposed model using Sentinel-2 satellite imagery for the An Khe reservoir in Vietnam.To assess its performance,we also conducted comparative experiments with other related models,including SegNet_ConvLSTM and UNet_ConvLSTM,on the same dataset.The model performances were validated using k-fold cross-validation and ANOVA analysis.The experimental results demonstrate that the YOLOv9_ConvLSTM model outperforms the compared models.It has been seen that the proposed approach serves as a valuable tool for reservoir water level forecasting using satellite imagery that contributes to effective water resource management.展开更多
The integration of deep learning into smart grid operations addresses critical challenges in dynamic load forecasting and optimal dispatch amid increasing renewable energy penetration.This study proposes a hybrid LSTM...The integration of deep learning into smart grid operations addresses critical challenges in dynamic load forecasting and optimal dispatch amid increasing renewable energy penetration.This study proposes a hybrid LSTM-Transformer architecture for multi-scale temporal-spatial load prediction,achieving 28%RMSE reduction on real-world datasets(CAISO,PJM),coupled with a deep reinforcement learning framework for multi-objective dispatch optimization that lowers operational costs by 12.4%while ensuring stability constraints.The synergy between adaptive forecasting models and scenario-based stochastic optimization demonstrates superior performance in handling renewable intermittency and demand volatility,validated through grid-scale case studies.Methodological innovations in federated feature extraction and carbon-aware scheduling further enhance scalability for distributed energy systems.These advancements provide actionable insights for grid operators transitioning to low-carbon paradigms,emphasizing computational efficiency and interoperability with legacy infrastructure.展开更多
Existing power forecasting models struggle to simultaneously handle high-dimensional,noisy load data while capturing long-term dependencies.This critical limitation necessitates an integrated approach combining dimens...Existing power forecasting models struggle to simultaneously handle high-dimensional,noisy load data while capturing long-term dependencies.This critical limitation necessitates an integrated approach combining dimensionality reduction,temporal modeling,and robust prediction,especially for multi-day forecasting.A novel hybrid model,SLHS-TCN-XGBoost,is proposed for power demand forecasting,leveraging SLHS(dimensionality reduction),TCN(temporal feature learning),and XGBoost(ensemble prediction).Applied to the three-year electricity load dataset of Seoul,South Korea,the model’s MAE,RMSE,and MAPE reached 112.08,148.39,and 2%,respectively,which are significantly reduced in MAE,RMSE,and MAPE by 87.37%,87.35%,and 87.43%relative to the baseline XGBoost model.Performance validation across nine forecast days demonstrates superior accuracy,with MAPE as low as 0.35%and 0.21%on key dates.Statistical Significance tests confirm significant improvements(p<0.05),with the highest MAPE reduction of 98.17%on critical days.Seasonal and temporal error analyses reveal stable performance,particularly in Quarter 3 and Quarter 4(0.5%,0.3%)and nighttime hours(<1%).Robustness tests,including 5-fold cross-validation and Various noise perturbations,confirm the model’s stability and resilience.The SLHS-TCN-XGBoost model offers an efficient and reliable solution for power demand forecasting,with future optimization potential in data preprocessing,algorithm integration,and interpretability.展开更多
To fully explore the potential features contained in power load data,an innovative short-term power load forecasting method that integrates data mining and deep learning techniques is proposed.Firstly,a density peak f...To fully explore the potential features contained in power load data,an innovative short-term power load forecasting method that integrates data mining and deep learning techniques is proposed.Firstly,a density peak fast search algorithm optimized by time series weighting factors is used to cluster and analyze load data,accurately dividing subsets of data into different categories.Secondly,introducing convolutional block attention mechanism into the bidirectional gated recurrent unit(BiGRU)structure significantly enhances its ability to extract key features.On this basis,in order to make the model more accurately adapt to the dynamic changes in power load data,subsets of different categories of data were used for BiGRU training based on attention mechanism,and extreme gradient boosting was selected as the meta model to effectively integrate multiple sets of historical training information.To further optimize the parameter configuration of the meta model,Bayesian optimization techniques are used to achieve automated adjustment of hyperparameters.Multiple sets of comparative experiments were designed,and the results showed that the average absolute error of the method in this paper was reduced by about 8.33%and 4.28%,respectively,compared with the single model and the combined model,and the determination coefficient reached the highest of 95.99,which proved that the proposed method has a better prediction effect.展开更多
Wind power forecasting plays a crucial role in optimizing the integration of wind energy into the grid by predicting wind patterns and energy output.This enhances the efficiency and reliability of renewable energy sys...Wind power forecasting plays a crucial role in optimizing the integration of wind energy into the grid by predicting wind patterns and energy output.This enhances the efficiency and reliability of renewable energy systems.Forecasting approaches inform energy management strategies,reduce reliance on fossil fuels,and support the broader transition to sustainable energy solutions.The primary goal of this study is to introduce an effective methodology for estimating wind power through temporal data analysis.This research advances an optimized Multilayer Perceptron(MLP)model using recently proposedmetaheuristic optimization algorithms,namely the FireHawk Optimizer(FHO)and the Non-Monopolize Search(NO).A modified version of FHO,termed FHONO,is developed by integrating NO as a local search mechanism to enhance the exploration capability and address the shortcomings of the original FHO.The developed FHONO is then employed to optimize the MLP for enhanced wind power prediction.The effectiveness of the proposed FHONO-MLP model is validated using renowned datasets from wind turbines in France.The results of the comparative analysis between FHONO-MLP,conventionalMLP,and other optimized versions of MLP show that FHONO-MLP outperforms the others,achieving an average RootMean Square Error(RMSE)of 0.105,Mean Absolute Error(MAE)of 0.082,and Coefficient of Determination(R^(2))of 0.967 across all datasets.These findings underscore the significant enhancement in predictive accuracy provided by FHONO and demonstrate its effectiveness in improving wind power forecasting.展开更多
Accurate subseasonal forecasting of East Asian summer monsoon(EASM)precipitation is crucial,as it directly impacts the livelihoods of billions.However,the prediction skill of state-of-the-art subseasonal-to-seasonal(S...Accurate subseasonal forecasting of East Asian summer monsoon(EASM)precipitation is crucial,as it directly impacts the livelihoods of billions.However,the prediction skill of state-of-the-art subseasonal-to-seasonal(S2S)models for precipitation remains limited.In this study,the authors developed a convolutional neural network(CNN)regression model to enhance the prediction skill for weekly EASM precipitation by utilizing the more reliably predicted circulation fields from dynamic models.The outcomes of the CNN model are promising,as it led to a 14%increase in the anomaly correlation coefficient(ACC),from 0.30 to 0.35,and a 22%reduction in the root-mean-square error(RMSE),from 3.22 to 2.52,for predicting the weekly EASM precipitation index at a leading time of one week.Among the S2S models,the improvement in prediction skill through CNN correction depends on the model’s performance in accurately predicting circulation fields.The CNN correction of EASM precipitation index can only rectify the systematic errors of the model and is independent of whether the each grid point or the entire area-averaged index is corrected.Furthermore,u200(200-hPa zonal wind)is identified as the most important variable for efficient correction.展开更多
Orthogonal conditional nonlinear optimal perturbations(O-CNOPs)have been used to generate ensemble forecasting members for achieving high forecasting skill of high-impact weather and climate events.However,highly effi...Orthogonal conditional nonlinear optimal perturbations(O-CNOPs)have been used to generate ensemble forecasting members for achieving high forecasting skill of high-impact weather and climate events.However,highly efficient calculations for O-CNOPs are still challenging in the field of ensemble forecasting.In this study,we combine a gradient-based iterative idea with the Gram‒Schmidt orthogonalization,and propose an iterative optimization method to compute O-CNOPs.This method is different from the original sequential optimization method,and allows parallel computations of O-CNOPs,thus saving a large amount of computational time.We evaluate this method by using the Lorenz-96 model on the basis of the ensemble forecasting ability achieved and on the time consumed for computing O-CNOPs.The results demonstrate that the parallel iterative method causes O-CNOPs to yield reliable ensemble members and to achieve ensemble forecasting skills similar to or even slightly higher than those produced by the sequential method.Moreover,the parallel method significantly reduces the computational time for O-CNOPs.Therefore,the parallel iterative method provides a highly effective and efficient approach for calculating O-CNOPs for ensemble forecasts.Expectedly,it can play an important role in the application of the O-CNOPs to realistic ensemble forecasts for high-impact weather and climate events.展开更多
Subseasonal-to-seasonal(S2S)forecasting for East Asian atmospheric circulation poses significant challenges for conventional numerical weather prediction(NWP)models.Recently,deep learning(DL)models have demonstrated s...Subseasonal-to-seasonal(S2S)forecasting for East Asian atmospheric circulation poses significant challenges for conventional numerical weather prediction(NWP)models.Recently,deep learning(DL)models have demonstrated significant potential in further enhancing S2S forecasts beyond the capabilities of NWP models.However,most current DLbased S2S forecasting models largely overlook the role of global predictors from multiple spheres,such as ocean,land,and atmosphere domains,that are crucial for effective S2S forecasting.In this study,we introduce EAAC-S2S,a tailored DL model for S2S forecasting of East Asian atmospheric circulation.EAAC-S2S employs the cross-attention mechanism to couple atmospheric circulations over East Asia with representative multi-sphere(i.e.,atmosphere,land,and ocean)variables,providing pentad-averaged circulation forecasts up to 12 pentads ahead throughout all seasons.Experimental results demonstrate,on the S2S time scale,that EAAC-S2S consistently outperforms the European Centre for MediumRange Weather Forecasts(ECMWF)Ensemble Prediction System by decreasing the root-mean-square error(RMSE)by3.8%and increasing the anomaly correlation coefficient(ACC)by 8.6%,averaged across all 17 predictands.Our system also shows good skill for examples of heatwaves and the South China Sea Subtropical High Intensity Index(SCSSHII).Moreover,quantitative interpretability analysis including multi-sphere attribution and attention visualization are conducted for the first time in a DL S2S model,where the traced predictability aligns well with prior meteorological knowledge.We hope that our results have the potential to advance research in data-driven S2S forecasting.展开更多
基金supported by the National Natural Science Foundation of China(Grant Nos.42375062 and 42275158)the National Key Scientific and Technological Infrastructure project“Earth System Science Numerical Simulator Facility”(EarthLab)the Natural Science Foundation of Gansu Province(Grant No.22JR5RF1080)。
文摘It is fundamental and useful to investigate how deep learning forecasting models(DLMs)perform compared to operational oceanography forecast systems(OFSs).However,few studies have intercompared their performances using an identical reference.In this study,three physically reasonable DLMs are implemented for the forecasting of the sea surface temperature(SST),sea level anomaly(SLA),and sea surface velocity in the South China Sea.The DLMs are validated against both the testing dataset and the“OceanPredict”Class 4 dataset.Results show that the DLMs'RMSEs against the latter increase by 44%,245%,302%,and 109%for SST,SLA,current speed,and direction,respectively,compared to those against the former.Therefore,different references have significant influences on the validation,and it is necessary to use an identical and independent reference to intercompare the DLMs and OFSs.Against the Class 4 dataset,the DLMs present significantly better performance for SLA than the OFSs,and slightly better performances for other variables.The error patterns of the DLMs and OFSs show a high degree of similarity,which is reasonable from the viewpoint of predictability,facilitating further applications of the DLMs.For extreme events,the DLMs and OFSs both present large but similar forecast errors for SLA and current speed,while the DLMs are likely to give larger errors for SST and current direction.This study provides an evaluation of the forecast skills of commonly used DLMs and provides an example to objectively intercompare different DLMs.
文摘Load forecasting is of great significance to the development of new power systems.With the advancement of smart grids,the integration and distribution of distributed renewable energy sources and power electronics devices have made power load data increasingly complex and volatile.This places higher demands on the prediction and analysis of power loads.In order to improve the prediction accuracy of short-term power load,a CNN-BiLSTMTPA short-term power prediction model based on the Improved Whale Optimization Algorithm(IWOA)with mixed strategies was proposed.Firstly,the model combined the Convolutional Neural Network(CNN)with the Bidirectional Long Short-Term Memory Network(BiLSTM)to fully extract the spatio-temporal characteristics of the load data itself.Then,the Temporal Pattern Attention(TPA)mechanism was introduced into the CNN-BiLSTM model to automatically assign corresponding weights to the hidden states of the BiLSTM.This allowed the model to differentiate the importance of load sequences at different time intervals.At the same time,in order to solve the problem of the difficulties of selecting the parameters of the temporal model,and the poor global search ability of the whale algorithm,which is easy to fall into the local optimization,the whale algorithm(IWOA)was optimized by using the hybrid strategy of Tent chaos mapping and Levy flight strategy,so as to better search the parameters of the model.In this experiment,the real load data of a region in Zhejiang was taken as an example to analyze,and the prediction accuracy(R2)of the proposed method reached 98.83%.Compared with the prediction models such as BP,WOA-CNN-BiLSTM,SSA-CNN-BiLSTM,CNN-BiGRU-Attention,etc.,the experimental results showed that the model proposed in this study has a higher prediction accuracy.
基金the Young Investigator Group“Artificial Intelligence for Probabilistic Weather Forecasting”funded by the Vector Stiftungfunding from the Federal Ministry of Education and Research(BMBF)and the Baden-Württemberg Ministry of Science as part of the Excellence Strategy of the German Federal and State Governments。
文摘Weather forecasts from numerical weather prediction models play a central role in solar energy forecasting,where a cascade of physics-based models is used in a model chain approach to convert forecasts of solar irradiance to solar power production.Ensemble simulations from such weather models aim to quantify uncertainty in the future development of the weather,and can be used to propagate this uncertainty through the model chain to generate probabilistic solar energy predictions.However,ensemble prediction systems are known to exhibit systematic errors,and thus require post-processing to obtain accurate and reliable probabilistic forecasts.The overarching aim of our study is to systematically evaluate different strategies to apply post-processing in model chain approaches with a specific focus on solar energy:not applying any post-processing at all;post-processing only the irradiance predictions before the conversion;post-processing only the solar power predictions obtained from the model chain;or applying post-processing in both steps.In a case study based on a benchmark dataset for the Jacumba solar plant in the U.S.,we develop statistical and machine learning methods for postprocessing ensemble predictions of global horizontal irradiance(GHI)and solar power generation.Further,we propose a neural-network-based model for direct solar power forecasting that bypasses the model chain.Our results indicate that postprocessing substantially improves the solar power generation forecasts,in particular when post-processing is applied to the power predictions.The machine learning methods for post-processing slightly outperform the statistical methods,and the direct forecasting approach performs comparably to the post-processing strategies.
文摘The China Meteorological Administration(CMA)said that in the last five years,China has made big improvements in its weather services.This includes better weather forecasts and ways to protect people from disasters.
基金supported by the open foundation of State Key Laboratory of Chemical Engineering(SKL-ChE-22B01)the Natural Science Foundation of China(22008169).
文摘The dynamics of vapor−liquid−solid(V−L−S)flow boiling in fluidized bed evaporators exhibit inherent complexity and chaotic behavior,hindering accurate prediction of pressure drop signals.To address this challenge,this study proposes an innovative hybrid approach that integrates wavelet neural network(WNN)with chaos analysis.By leveraging the Cross-Correlation(C−C)method,the minimum embedding dimension for phase space reconstruction is systematically calculated and then adopted as the input node configuration for the WNN.Simulation results demonstrate the remarkable effectiveness of this integrated method in predicting pressure drop signals,advancing our understanding of the intricate dynamic phenomena occurring with V−L−S fluidized bed evaporators.Moreover,this study offers a novel perspective on applying advanced data-driven techniques to handle the complexities of multi-phase flow systems and highlights the potential for improved operational prediction and control in industrial settings.
文摘Time series forecasting is important in the fields of finance,energy,and meteorology,but traditional methods often fail to cope with the complex nonlinear and nonstationary processes of real data.In this paper,we propose the FractalNet-LSTM model,which combines fractal convolutional units with recurrent long short-term memory(LSTM)layers to model time series efficiently.To test the effectiveness of the model,data with complex structures and patterns,in particular,with seasonal and cyclical effects,were used.To better demonstrate the obtained results and the formed conclusions,the model performance was shown on the datasets of electricity consumption,sunspot activity,and Spotify stock price.The result showed that the proposed model outperforms traditional approaches at medium forecasting horizons and demonstrates high accuracy for data with long-term and cyclical dependencies.However,for financial data with high volatility,the model’s efficiency decreases at long forecasting horizons,indicating the need for further adaptation.The findings suggest further adaptation.The findings suggest that integrating fractal properties into neural network architecture improves the accuracy of time series forecasting and can be useful for developing more accurate and reliable forecasting systems in various industries.
基金jointly supported by the Innovation Group Project of the Southern Marine Science and Engineering Guangdong Laboratory(Zhuhai)[grant number 316323005]the Guangdong Basic and Applied Basic Research Foundation[grant numbers 2023A1515010741 and 2024B1515020035]the Science and Technology Planning Project of Guangdong Province[grant number 2023B1212060019]。
文摘Forecasting tropical cyclone(TC)activities has been a topic of great interest and research.Taiwan Island(TW)is one of the key regions that is highly exposed to TCs originated from the western North Pacific.Here,the authors utilize two mainstream reanalysis datasets for the period 1979-2013 and propose an effective statistical seasonal forecasting model-namely,the Sun Yat-sen University(SYSU)Model-for predicting the number of TC landfalls on TW based on the environmental factors in the preseason.The comprehensive predictor sampling and multiple linear regression show that the 850-hPa meridional wind over the west of the Antarctic Peninsula in January,the 300-hPa specific humidity over the open ocean southwest of Australia in January,the 300-hPa relative vorticity over the west of the Sea of Okhotsk in March,and the sea surface temperature in the South Indian Ocean in April,are the most significant predictors.The correlation coefficient between the modeled results and observations reaches 0.87.The model is validated by the leave-one-out and nine-fold cross-validation methods,and recent 9-yr observations(2014-2022).The Antarctic Oscillation,variabilities of the western Pacific subtropical high,Asian summer monsoon,and oceanic tunnel are the possible physical linkages or mechanisms behind the model result.The SYSU Model exhibits a 98%hit rate in 1979-2022(43 out of 44),suggesting an operational potential in the seasonal forecasting of TC landfalls on TW.
文摘Based on ground observation data of relative humidity,the prediction performance of STNF and MIFS in each competition area during February 13-26,2024 was tested and evaluated by using two intelligent forecasting methods(STNF and MIFS).The results show that STNF had better performance in forecasting relative humidity in high-altitude areas,and was suitable for fine forecasting under complex terrain.MIFS improved the short-term forecast of some low-altitude stations,but the long-term reliability was insufficient.STNF method performed better than MIFS during 0-24 h.As the prediction time extended to 24-72 h,the errors of both methods showed a systematic increase trend.STNF had higher precision,lower root mean square error and smaller mean error in most regions under the background of most weather systems,showing its superiority as a forecasting method of relative humidity.However,the precision of MIFS was slightly higher than that of STNF in Liangcheng without system background,revealing that MIFS may also be an effective option in some specific conditions.
基金supported by the Startup Grant(PG18929)awarded to F.Shokoohi.
文摘Accurate Electric Load Forecasting(ELF)is crucial for optimizing production capacity,improving operational efficiency,and managing energy resources effectively.Moreover,precise ELF contributes to a smaller environmental footprint by reducing the risks of disruption,downtime,and waste.However,with increasingly complex energy consumption patterns driven by renewable energy integration and changing consumer behaviors,no single approach has emerged as universally effective.In response,this research presents a hybrid modeling framework that combines the strengths of Random Forest(RF)and Autoregressive Integrated Moving Average(ARIMA)models,enhanced with advanced feature selection—Minimum Redundancy Maximum Relevancy and Maximum Synergy(MRMRMS)method—to produce a sparse model.Additionally,the residual patterns are analyzed to enhance forecast accuracy.High-resolution weather data from Weather Underground and historical energy consumption data from PJM for Duke Energy Ohio and Kentucky(DEO&K)are used in this application.This methodology,termed SP-RF-ARIMA,is evaluated against existing approaches;it demonstrates more than 40%reduction in mean absolute error and root mean square error compared to the second-best method.
基金supported in part by the Meteorological Joint Funds of the National Natural Science Foundation of China under Grant U2142211in part by the National Natural Science Foundation of China under Grant 42075141,42341202+2 种基金in part by the National Key Research and Development Program of China under Grant 2020YFA0608000in part by the Shanghai Municipal Science and Technology Major Project(2021SHZDZX0100)the Fundamental Research Funds for the Central Universities。
文摘In this paper,we introduce TianXing,a transformer-based data-driven model designed with physical augmentation for skillful and efficient global weather forecasting.Previous data-driven transformer models such as Pangu-Weather,FengWu,and FuXi have emerged as promising alternatives for numerical weather prediction in weather forecasting.However,these models have been characterized by their substantial computational resource consumption during training and limited incorporation of explicit physical guidance in their modeling frameworks.In contrast,TianXing applies a linear complexity mechanism that ensures proportional scalability with input data size while significantly diminishing GPU resource demands,with only a marginal compromise in accuracy.Furthermore,TianXing proposes an explicit attention decay mechanism in the linear attention derived from physical insights to enhance its forecasting skill.The mechanism can reweight attention based on Earth's spherical distances and learned sparse multivariate coupling relationships,promptingTianXing to prioritize dynamically relevant neighboring features.Finally,to enhance its performance in mediumrange forecasting,TianXing employs a stacked autoregressive forecast algorithm.Validation of the model's architecture is conducted using ERA5 reanalysis data at a 5.625°latitude-longitude resolution,while a high-resolution dataset at 0.25°is utilized for training the actual forecasting model.Notably,the TianXing exhibits excellent performance,particularly in the Z500(geopotential height)and T850(temperature)fields,surpassing previous data-driven models and operational fullresolution models such as NCEP GFS and ECMWF IFS,as evidenced by latitude-weighted RMSE and ACC metrics.Moreover,the TianXing has demonstrated remarkable capabilities in predicting extreme weather events,such as typhoons.
基金funded by Natural Science Foundation of Heilongjiang Province,grant number LH2023F020.
文摘Deep learning(DL)has revolutionized time series forecasting(TSF),surpassing traditional statistical methods(e.g.,ARIMA)and machine learning techniques in modeling complex nonlinear dynamics and long-term dependencies prevalent in real-world temporal data.This comprehensive survey reviews state-of-the-art DL architectures forTSF,focusing on four core paradigms:(1)ConvolutionalNeuralNetworks(CNNs),adept at extracting localized temporal features;(2)Recurrent Neural Networks(RNNs)and their advanced variants(LSTM,GRU),designed for sequential dependency modeling;(3)Graph Neural Networks(GNNs),specialized for forecasting structured relational data with spatial-temporal dependencies;and(4)Transformer-based models,leveraging self-attention mechanisms to capture global temporal patterns efficiently.We provide a rigorous analysis of the theoretical underpinnings,recent algorithmic advancements(e.g.,TCNs,attention mechanisms,hybrid architectures),and practical applications of each framework,supported by extensive benchmark datasets(e.g.,ETT,traffic flow,financial indicators)and standardized evaluation metrics(MAE,MSE,RMSE).Critical challenges,including handling irregular sampling intervals,integrating domain knowledge for robustness,and managing computational complexity,are thoroughly discussed.Emerging research directions highlighted include diffusion models for uncertainty quantification,hybrid pipelines combining classical statistical and DL techniques for enhanced interpretability,quantile regression with Transformers for riskaware forecasting,and optimizations for real-time deployment.This work serves as an essential reference,consolidating methodological innovations,empirical resources,and future trends to bridge the gap between theoretical research and practical implementation needs for researchers and practitioners in the field.
基金supported by Key Laboratory of Smart Earth(KF2023ZD03-05)CMA Innovative and Development Program(CXFZ.20231035)+2 种基金National Key R&D Program of China(No.2021ZD0111902)National Natural Science Foundation of China(Nos.62472014,U21B2038)the Scientific and Technological Project of China Meteorological Administration(CMAJBGS202505).
文摘Weather forecasting is crucial for agriculture,transportation,and industry.Deep Learning(DL)has greatly improved the prediction accuracy.Among them,Graph Neural Networks(GNNs)excel at processing weather data by establishing connections between regions.This allows them to understand complex patterns that traditional methods might miss.As a result,achieving more accurate predictions becomes possible.The paper reviews the role of GNNs in short-to medium-range weather forecasting.The methods are classified into three categories based on dataset differences.The paper also further identifies five promising research frontiers.These areas aim to boost forecasting precision and enhance computational efficiency.They offer valuable insights for future weather forecasting systems.
基金funded by International School,Vietnam National University,Hanoi(VNU-IS)under project number CS.2023-10.
文摘Global climate change,along with the rapid increase of the population,has put significant pressure on water security.A water reservoir is an effective solution for adjusting and ensuring water supply.In particular,the reservoir water level is an essential physical indicator for the reservoirs.Forecasting the reservoir water level effectively assists the managers in making decisions and plans related to reservoir management policies.In recent years,deep learning models have been widely applied to solve forecasting problems.In this study,we propose a novel hybrid deep learning model namely the YOLOv9_ConvLSTM that integrates YOLOv9,ConvLSTM,and linear interpolation to predict reservoir water levels.It utilizes data from Sentinel-2 satellite images,generated from visible spectrum bands(Red-Blue-Green)to reconstruct true-color reservoir images.Adam is used as the optimization algorithm with the loss function being MSE(Mean Squared Error)to evaluate the model’s error during training.We implemented and validated the proposed model using Sentinel-2 satellite imagery for the An Khe reservoir in Vietnam.To assess its performance,we also conducted comparative experiments with other related models,including SegNet_ConvLSTM and UNet_ConvLSTM,on the same dataset.The model performances were validated using k-fold cross-validation and ANOVA analysis.The experimental results demonstrate that the YOLOv9_ConvLSTM model outperforms the compared models.It has been seen that the proposed approach serves as a valuable tool for reservoir water level forecasting using satellite imagery that contributes to effective water resource management.
文摘The integration of deep learning into smart grid operations addresses critical challenges in dynamic load forecasting and optimal dispatch amid increasing renewable energy penetration.This study proposes a hybrid LSTM-Transformer architecture for multi-scale temporal-spatial load prediction,achieving 28%RMSE reduction on real-world datasets(CAISO,PJM),coupled with a deep reinforcement learning framework for multi-objective dispatch optimization that lowers operational costs by 12.4%while ensuring stability constraints.The synergy between adaptive forecasting models and scenario-based stochastic optimization demonstrates superior performance in handling renewable intermittency and demand volatility,validated through grid-scale case studies.Methodological innovations in federated feature extraction and carbon-aware scheduling further enhance scalability for distributed energy systems.These advancements provide actionable insights for grid operators transitioning to low-carbon paradigms,emphasizing computational efficiency and interoperability with legacy infrastructure.
基金supported by Mahasarakham University for Piyapatr Busababodhin’s work.Guoqing Chen’s research was supported by Chengdu Jincheng College Green Data Integration Intelligence Research and Innovation Project(No.2025-2027)the High-Quality Development Research Center Project in the Tuojiang River Basin(No.TJGZL2024-07)+1 种基金the Open Fund ofWuhan Gravitation and Solid Earth Tides,National Observation and Research Station(No.WHYWZ202406)the Scientific Research Fund of the Institute of Seismology,CEA,and the National Institute of Natural Hazards,MEM(No.IS202236328).
文摘Existing power forecasting models struggle to simultaneously handle high-dimensional,noisy load data while capturing long-term dependencies.This critical limitation necessitates an integrated approach combining dimensionality reduction,temporal modeling,and robust prediction,especially for multi-day forecasting.A novel hybrid model,SLHS-TCN-XGBoost,is proposed for power demand forecasting,leveraging SLHS(dimensionality reduction),TCN(temporal feature learning),and XGBoost(ensemble prediction).Applied to the three-year electricity load dataset of Seoul,South Korea,the model’s MAE,RMSE,and MAPE reached 112.08,148.39,and 2%,respectively,which are significantly reduced in MAE,RMSE,and MAPE by 87.37%,87.35%,and 87.43%relative to the baseline XGBoost model.Performance validation across nine forecast days demonstrates superior accuracy,with MAPE as low as 0.35%and 0.21%on key dates.Statistical Significance tests confirm significant improvements(p<0.05),with the highest MAPE reduction of 98.17%on critical days.Seasonal and temporal error analyses reveal stable performance,particularly in Quarter 3 and Quarter 4(0.5%,0.3%)and nighttime hours(<1%).Robustness tests,including 5-fold cross-validation and Various noise perturbations,confirm the model’s stability and resilience.The SLHS-TCN-XGBoost model offers an efficient and reliable solution for power demand forecasting,with future optimization potential in data preprocessing,algorithm integration,and interpretability.
基金supported in part by the Fundamental Research Funds for the Liaoning Universities(LJ212410146025)the Graduate Science and Technology Innovation Project of University of Science and Technology Liaoning(LKDYC202310).
文摘To fully explore the potential features contained in power load data,an innovative short-term power load forecasting method that integrates data mining and deep learning techniques is proposed.Firstly,a density peak fast search algorithm optimized by time series weighting factors is used to cluster and analyze load data,accurately dividing subsets of data into different categories.Secondly,introducing convolutional block attention mechanism into the bidirectional gated recurrent unit(BiGRU)structure significantly enhances its ability to extract key features.On this basis,in order to make the model more accurately adapt to the dynamic changes in power load data,subsets of different categories of data were used for BiGRU training based on attention mechanism,and extreme gradient boosting was selected as the meta model to effectively integrate multiple sets of historical training information.To further optimize the parameter configuration of the meta model,Bayesian optimization techniques are used to achieve automated adjustment of hyperparameters.Multiple sets of comparative experiments were designed,and the results showed that the average absolute error of the method in this paper was reduced by about 8.33%and 4.28%,respectively,compared with the single model and the combined model,and the determination coefficient reached the highest of 95.99,which proved that the proposed method has a better prediction effect.
基金the Deanship of Graduate Studies and Scientific Research at University of Bisha,Saudi Arabia for funding this research work through the Promising Program under Grant Number(UB-Promising-42-1445).
文摘Wind power forecasting plays a crucial role in optimizing the integration of wind energy into the grid by predicting wind patterns and energy output.This enhances the efficiency and reliability of renewable energy systems.Forecasting approaches inform energy management strategies,reduce reliance on fossil fuels,and support the broader transition to sustainable energy solutions.The primary goal of this study is to introduce an effective methodology for estimating wind power through temporal data analysis.This research advances an optimized Multilayer Perceptron(MLP)model using recently proposedmetaheuristic optimization algorithms,namely the FireHawk Optimizer(FHO)and the Non-Monopolize Search(NO).A modified version of FHO,termed FHONO,is developed by integrating NO as a local search mechanism to enhance the exploration capability and address the shortcomings of the original FHO.The developed FHONO is then employed to optimize the MLP for enhanced wind power prediction.The effectiveness of the proposed FHONO-MLP model is validated using renowned datasets from wind turbines in France.The results of the comparative analysis between FHONO-MLP,conventionalMLP,and other optimized versions of MLP show that FHONO-MLP outperforms the others,achieving an average RootMean Square Error(RMSE)of 0.105,Mean Absolute Error(MAE)of 0.082,and Coefficient of Determination(R^(2))of 0.967 across all datasets.These findings underscore the significant enhancement in predictive accuracy provided by FHONO and demonstrate its effectiveness in improving wind power forecasting.
基金supported by a Guangdong Major Project of Basic and Applied Basic Research[grant number 2020B0301030004]the National Natural Science Foundation of China[grant number 42175061]。
文摘Accurate subseasonal forecasting of East Asian summer monsoon(EASM)precipitation is crucial,as it directly impacts the livelihoods of billions.However,the prediction skill of state-of-the-art subseasonal-to-seasonal(S2S)models for precipitation remains limited.In this study,the authors developed a convolutional neural network(CNN)regression model to enhance the prediction skill for weekly EASM precipitation by utilizing the more reliably predicted circulation fields from dynamic models.The outcomes of the CNN model are promising,as it led to a 14%increase in the anomaly correlation coefficient(ACC),from 0.30 to 0.35,and a 22%reduction in the root-mean-square error(RMSE),from 3.22 to 2.52,for predicting the weekly EASM precipitation index at a leading time of one week.Among the S2S models,the improvement in prediction skill through CNN correction depends on the model’s performance in accurately predicting circulation fields.The CNN correction of EASM precipitation index can only rectify the systematic errors of the model and is independent of whether the each grid point or the entire area-averaged index is corrected.Furthermore,u200(200-hPa zonal wind)is identified as the most important variable for efficient correction.
基金sponsored by the National Natural Science Foundation of China(Grant Nos.41930971,42330111,and 42405061)the National Key Scientific and Technological Infrastructure project“Earth System Numerical Simulation Facility”(Earth Lab).
文摘Orthogonal conditional nonlinear optimal perturbations(O-CNOPs)have been used to generate ensemble forecasting members for achieving high forecasting skill of high-impact weather and climate events.However,highly efficient calculations for O-CNOPs are still challenging in the field of ensemble forecasting.In this study,we combine a gradient-based iterative idea with the Gram‒Schmidt orthogonalization,and propose an iterative optimization method to compute O-CNOPs.This method is different from the original sequential optimization method,and allows parallel computations of O-CNOPs,thus saving a large amount of computational time.We evaluate this method by using the Lorenz-96 model on the basis of the ensemble forecasting ability achieved and on the time consumed for computing O-CNOPs.The results demonstrate that the parallel iterative method causes O-CNOPs to yield reliable ensemble members and to achieve ensemble forecasting skills similar to or even slightly higher than those produced by the sequential method.Moreover,the parallel method significantly reduces the computational time for O-CNOPs.Therefore,the parallel iterative method provides a highly effective and efficient approach for calculating O-CNOPs for ensemble forecasts.Expectedly,it can play an important role in the application of the O-CNOPs to realistic ensemble forecasts for high-impact weather and climate events.
基金supported in part by the Meteorological Joint Funds of the National Natural Science Foundation of China(Grant No.U2142211)by the National Key Research and Development Program of China(Grant No.2020YFA0608002)+4 种基金by the National Natural Science Foundation of China(Grant Nos.42075141 and 42341202)by the China National Postdoctoral Program for Innovative Talents(Grant No.BX20230071)by the National Natural Science Foundation of China for Youth(Grant No.42205191)by the Shanghai Municipal Science and Technology Major Project(Grant No.2021SHZDZX0100)the Fundamental Research Funds for the Central Universities。
文摘Subseasonal-to-seasonal(S2S)forecasting for East Asian atmospheric circulation poses significant challenges for conventional numerical weather prediction(NWP)models.Recently,deep learning(DL)models have demonstrated significant potential in further enhancing S2S forecasts beyond the capabilities of NWP models.However,most current DLbased S2S forecasting models largely overlook the role of global predictors from multiple spheres,such as ocean,land,and atmosphere domains,that are crucial for effective S2S forecasting.In this study,we introduce EAAC-S2S,a tailored DL model for S2S forecasting of East Asian atmospheric circulation.EAAC-S2S employs the cross-attention mechanism to couple atmospheric circulations over East Asia with representative multi-sphere(i.e.,atmosphere,land,and ocean)variables,providing pentad-averaged circulation forecasts up to 12 pentads ahead throughout all seasons.Experimental results demonstrate,on the S2S time scale,that EAAC-S2S consistently outperforms the European Centre for MediumRange Weather Forecasts(ECMWF)Ensemble Prediction System by decreasing the root-mean-square error(RMSE)by3.8%and increasing the anomaly correlation coefficient(ACC)by 8.6%,averaged across all 17 predictands.Our system also shows good skill for examples of heatwaves and the South China Sea Subtropical High Intensity Index(SCSSHII).Moreover,quantitative interpretability analysis including multi-sphere attribution and attention visualization are conducted for the first time in a DL S2S model,where the traced predictability aligns well with prior meteorological knowledge.We hope that our results have the potential to advance research in data-driven S2S forecasting.
