This study investigates multi-model ensemble forecasts of track and intensity of tropical cyclones over the western Pacific, based on forecast outputs from the China Meteorological Administration, European Centre for ...This study investigates multi-model ensemble forecasts of track and intensity of tropical cyclones over the western Pacific, based on forecast outputs from the China Meteorological Administration, European Centre for Medium-Range Weather Forecasts, Japan Meteorological Agency and National Centers for Environmental Prediction in the THORPEX Interactive Grand Global Ensemble(TIGGE) datasets. The multi-model ensemble schemes, namely the bias-removed ensemble mean(BREM) and superensemble(SUP), are compared with the ensemble mean(EMN) and single-model forecasts. Moreover, a new model bias estimation scheme is investigated and applied to the BREM and SUP schemes. The results showed that, compared with single-model forecasts and EMN, the multi-model ensembles of the BREM and SUP schemes can have smaller errors in most cases. However, there were also circumstances where BREM was less skillful than EMN, indicating that using a time-averaged error as model bias is not optimal. A new model bias estimation scheme of the biweight mean is introduced. Through minimizing the negative influence of singular errors, this scheme can obtain a more accurate model bias estimation and improve the BREM forecast skill. The application of the biweight mean in the bias calculation of SUP also resulted in improved skill. The results indicate that the modification of multi-model ensemble schemes through this bias estimation method is feasible.展开更多
This study focuses on an extreme rainfall event in East China during the mei-yu season,in which the capital city(Nanjing)of Jiangsu Province experienced a maximum 14-h rainfall accumulation of 209.6 mm and a peak hour...This study focuses on an extreme rainfall event in East China during the mei-yu season,in which the capital city(Nanjing)of Jiangsu Province experienced a maximum 14-h rainfall accumulation of 209.6 mm and a peak hourly rainfall of 118.8 mm.The performance of two sets of convection-permitting ensemble forecast systems(CEFSs),each with 30 members and a 3-km horizontal grid spacing,is evaluated.The CEFS_ICBCs,using multiple initial and boundary conditions(ICs and BCs),and the CEFS_ICBCs Phys,which incorporates both multi-physics schemes and ICs/BCs,are compared to the CMA-REPS(China Meteorological Administration-Regional Ensemble Prediction System)with a coarser 10-km grid spacing.The two CEFSs demonstrate more uniform rank histograms and lower Brier scores(with higher resolution),improving precipitation intensity predictions and providing more reliable probability forecasts,although they overestimate precipitation over Mt.Dabie.It is challenging for the CEFSs to capture the evolution of mesoscale rainstorms that are known to be related to the errors in predicting the southwesterly low-level winds.Sensitivity experiments reveal that the microphysics and radiation schemes introduce considerable uncertainty in predicting the intensity and location of heavy rainfall in and near Nanjing and Mt.Dabie.In particular,the Asymmetric Convection Model 2(ACM2)planetary boundary layer scheme combined with the Pleim-Xiu surface layer scheme tends to produce a biased northeastward extension of the boundary-layer jet,contributing to the northeastward bias of heavy precipitation around Nanjing in the CEFS_ICBCs.展开更多
Grid forecasting can be used to effectively enhance the spatial and temporal density of forecast products,thereby improving the capability of short-term marine disaster forecasting and warnings in terms of proximity.T...Grid forecasting can be used to effectively enhance the spatial and temporal density of forecast products,thereby improving the capability of short-term marine disaster forecasting and warnings in terms of proximity.The traditional method that relies on forecasters'subjective correction of station observation data for forecasting has been unable to meet the practical needs of refined forecasting.To address this problem,this paper proposes a Transformer-enhanced UNet(TransUNet)model for wave forecast AI correction,which fuses wind and wave information.The Transformer structure is integrated into the encoder of the UNet model,and instead of using the traditional upsampling method,the dual-sampling module is employed in the decoder to enhance the feature extraction capability.This paper compares the TransUNet model with the traditional UNet model using wind speed forecast data,wave height forecast data,and significant wave height reanalysis data provided by ECMWF.The experimental results indicate that the TransUNet model yields smaller root-meansquare errors,mean errors,and standard deviations of the corrected results for the next 24-h forecasts than does the UNet model.Specifically,the root-mean-square error decreased by more than 21.55%compared to its precorrection value.According to the statistical analysis,87.81%of the corrected wave height errors for the next 24-h forecast were within±0.2m,with only 4.56%falling beyond±0.3 m.This model effectively limits the error range and enhances the ability to forecast wave heights.展开更多
This study investigates the load-bearing capacity of open-ended pipe piles in sandy soil, with a specific focus on the impact of soil plug constraints at four levels(no plug, 25% plug, 50% plug, and full plug). Levera...This study investigates the load-bearing capacity of open-ended pipe piles in sandy soil, with a specific focus on the impact of soil plug constraints at four levels(no plug, 25% plug, 50% plug, and full plug). Leveraging a dataset comprising open-ended pipe piles with varying geometrical and geotechnical properties, this research employs shallow neural network(SNN) and deep neural network(DNN) models to predict plugging conditions for both driven and pressed installation types. This paper underscores the importance of key parameters such as the settlement value,applied load, installation type, and soil configuration(loose, medium, and dense) in accurately predicting pile settlement. These findings offer valuable insights for optimizing pile design and construction in geotechnical engineering,addressing a longstanding challenge in the field. The study demonstrates the potential of the SNN and DNN models in precisely identifying plugging conditions before pile driving, with the SNN achieving R2 values ranging from0.444 to 0.711 and RMSPE values ranging from 24.621% to 48.663%, whereas the DNN exhibits superior performance, with R2 values ranging from 0.815 to 0.942 and RMSPE values ranging from 4.419% to 10.325%. These results have significant implications for enhancing construction practices and reducing uncertainties associated with pile foundation projects in addition to leveraging artificial intelligence tools to avoid long experimental procedures.展开更多
The impacts of lateral boundary conditions(LBCs)provided by numerical models and data-driven networks on convective-scale ensemble forecasts are investigated in this study.Four experiments are conducted on the Hangzho...The impacts of lateral boundary conditions(LBCs)provided by numerical models and data-driven networks on convective-scale ensemble forecasts are investigated in this study.Four experiments are conducted on the Hangzhou RDP(19th Hangzhou Asian Games Research Development Project on Convective-scale Ensemble Prediction and Application)testbed,with the LBCs respectively sourced from National Centers for Environmental Prediction(NCEP)Global Forecast System(GFS)forecasts with 33 vertical levels(Exp_GFS),Pangu forecasts with 13 vertical levels(Exp_Pangu),Fuxi forecasts with 13 vertical levels(Exp_Fuxi),and NCEP GFS forecasts with the vertical levels reduced to 13(the same as those of Exp_Pangu and Exp_Fuxi)(Exp_GFSRDV).In general,Exp_Pangu performs comparably to Exp_GFS,while Exp_Fuxi shows slightly inferior performance compared to Exp_Pangu,possibly due to its less accurate large-scale predictions.Therefore,the ability of using data-driven networks to efficiently provide LBCs for convective-scale ensemble forecasts has been demonstrated.Moreover,Exp_GFSRDV has the worst convective-scale forecasts among the four experiments,which indicates the potential improvement of using data-driven networks for LBCs by increasing the vertical levels of the networks.However,the ensemble spread of the four experiments barely increases with lead time.Thus,each experiment has insufficient ensemble spread to present realistic forecast uncertainties,which will be investigated in a future study.展开更多
The financial health of leading enterprises has a significant impact on the sustainable development of the global economy.Most data-driven financial health forecasts are based on the direct use of small-scale machine ...The financial health of leading enterprises has a significant impact on the sustainable development of the global economy.Most data-driven financial health forecasts are based on the direct use of small-scale machine learning.In this study,we proposed the idea of optimization coupling learning to improve these machine learning models in financial health forecasting.It not only revealed lagging,immediate,continuous impacts of various indicators in different fiscal year,but also had the same low computational cost and complexity as known small-scale machine learning models.We used our optimization coupling learning to investigate 3424 leading enterprises in China and revealed inner triggering mechanisms and differences of enterprises’financial health status from individual behavior to macro level.展开更多
Moisture conditions are crucial for the maintenance and development of severe convection.In the indirect assimilation of radar reflectivity,hydrometeors and water vapor retrieved from reflectivity are assimilated to a...Moisture conditions are crucial for the maintenance and development of severe convection.In the indirect assimilation of radar reflectivity,hydrometeors and water vapor retrieved from reflectivity are assimilated to avoid the nonlinearity issues associated with the observation operator.In a widely applied water vapor retrieval scheme,a cloud is assumed to be saturated when the radar reflectivity exceeds a certain threshold.This study replaces the traditional retrieval scheme with the“Z-RH”(radar reflectivity and relative humidity)linear statistical relationship for estimating the water vapor content,which is implemented to reduce the uncertainty caused by empirical relationships.The“Z-RH”relationship is statistically obtained from the humidity and the observations for rainfall rate at different temperature intervals with the use of the Z-R(radar reflectivity-rain rate)relationship.The impacts of these two retrieval approaches are investigated in the analyses and forecasts based on the radar reflectivity.The results suggest that both water vapor retrieval schemes yield similar reflectivity analyses,with“Z-RH”showing slightly stronger reflectivity intensities.Utilizing a“Z-RH”scheme contributes significantly to the improved analyses and forecasts of humidity and wind fields,resulting in more reasonable thermodynamic and dynamic structures.As the“Z-RH”relationship obtained by real-time statistics in a specific area provides a scientific basis for the retrieval of water vapor,a“Z-RH”scheme is beneficial to obtain more accurate reflectivity forecasts.The overall scores for the predicted precipitation of a“Z-RH”scheme are roughly 10%-20%higher compared to those of the traditional scheme.展开更多
Symposium overview The International Symposium on Earthquake Forecasting to Commemorate the 50th Anniversary of the 1975 Haicheng M7.3 Earthquake,Liaoning,China,was held in Shenyang,China,from 8 to 11 July 2025.The sy...Symposium overview The International Symposium on Earthquake Forecasting to Commemorate the 50th Anniversary of the 1975 Haicheng M7.3 Earthquake,Liaoning,China,was held in Shenyang,China,from 8 to 11 July 2025.The symposium was organized by the Institute of Earthquake Forecasting,China Earthquake Administration(CEA),the State Key Laboratory of Earthquake Dynamics and Forecasting,and the China Seismic Experimental Site(CSES),in collaboration with the International Association of Seismology and Physics of the Earth’s Interior(IASPEI),the APEC Cooperation for Earthquake Science(ACES).展开更多
OBJECTIVE:To develop an automated system for identifying and classifying constitution types in Traditional Chinese Medicine(TCM)by leveraging multi-model fusion algorithms.METHODS:A condensed version of a physical inf...OBJECTIVE:To develop an automated system for identifying and classifying constitution types in Traditional Chinese Medicine(TCM)by leveraging multi-model fusion algorithms.METHODS:A condensed version of a physical information collection form was designed to facilitate efficient data acquisition.The collected data were analyzed using a multi-model fusion approach,which integrated several machine learning techniques.These included support vector machines,Naive Bayes,decision trees,random forests,logistic regression,multilayer perceptrons,K-nearest neighbors,gradient boosting,adaptive ensemble learning,and recurrent neural networks.A soft voting strategy was used to combine the predictive outputs of each model,enabling the selection of the most effective model combination.RESULTS:The classification models demonstrated consistent and robust performance across most TCM constitution types when enhanced by the multi-model fusion strategy.In particular,high levels of accuracy,precision,recall,and F1-score were achieved for constitution types such as Yang deficiency,Qi deficiency,and Qi stagnation.However,the classification performance for the Yin deficiency constitution was relatively lower,indicating the need for further refinement and optimization in future research.CONCLUSION:This study introduces a novel,automated method for classifying TCM constitution types through the application of multi-model fusion algorithms.The approach simplifies the complex task of constitution identification while offering a practical and theoretical framework for the intelligent diagnosis of TCM body types.The findings have the potential to enhance personalized health management and support clinical decision-making in TCM diagnosis and treatment.展开更多
Since the initiation of the subseasonal-to-seasonal prediction project by the World Meteorological Organization,the accuracy of model forecasts has improved notably.However,substantial discrepancies have been observed...Since the initiation of the subseasonal-to-seasonal prediction project by the World Meteorological Organization,the accuracy of model forecasts has improved notably.However,substantial discrepancies have been observed among forecast results produced by different ensemble members when applied to South China.To enhance the accuracy of sub-seasonal forecasts in this region,it is essential to develop new methods that can effectively leverage multiple predictive models.This study introduces a weighted ensemble forecasting method based on online learning to improve forecast accuracy.We utilized ensemble forecasts from three models:the Integrated Forecasting System model from the European Centre for Medium-Range Weather Forecasts,the Climate Forecast System Version 2 model from the National Centers for Environmental Prediction,and the Beijing Climate Center-Climate Prediction System version 3 model from the China Meteorological Administration.The ensemble weights are trained using an online learning approach.The results indicate that the forecasts obtained through online learning outperform those of the original dynamical models.Compared to the simple ensemble results of the three models,the weighted ensemble model showed a stronger capability to capture temperature and precipitation patterns in South China.Therefore,this method has the potential to improve the accuracy of sub-seasonal forecasts in this region.展开更多
Wind speed is a crucial parameter affecting wind energy utilization.However,its volatility leads to time-varying power output.Herein,a novel Seq2Seq model integrating deep learning,data denoising,and a shape-aware los...Wind speed is a crucial parameter affecting wind energy utilization.However,its volatility leads to time-varying power output.Herein,a novel Seq2Seq model integrating deep learning,data denoising,and a shape-aware loss function is proposed for accurate multistep wind speed forecasting.In this model,the wind speed data is first denoised using the maximal overlap discrete wavelet transform.Next,an encoder-decoder network based on a temporal convolutional network,bidirectional gated recurrent unit,and multihead self-attention is employed for forecasting.Additionally,to enhance the ability of the model to identify temporal dynamics,a shape-aware loss function,ITILDE-Q,is employed in the model.To verify the effectiveness of the proposed model,a comparative experiment and an ablation experiment were conducted using three datasets of measured wind speeds.Three error metrics and a similarity metric were adopted for comprehensive evaluation.The experimental results showed that the proposed model consistently outperforms benchmark models in all tested forecasting scenarios,with particularly pronounced differences in performance over longer forecast horizons.Furthermore,the synergistic interaction of the three key components contributes to the extraordinary performance of the proposed model.展开更多
[Objectives]This study was conducted to achieve rapid and accurate detection of protein content in rice with a particle size of 1.0 mm.[Methods]A multi-model fusion strategy was proposed on the basis of Stacking ensem...[Objectives]This study was conducted to achieve rapid and accurate detection of protein content in rice with a particle size of 1.0 mm.[Methods]A multi-model fusion strategy was proposed on the basis of Stacking ensemble learning.A base learner pool was constructed,containing Partial Least Squares(PLS),Support Vector Machine(SVM),Deep Extreme Learning Machine(DELM),Random Forest(RF),Gradient Boosting Decision Tree(GBDT),and Multilayer Perceptron(MLP).PLS,DELM,and Linear Regression(LR)were used as meta-learner candidates.Employing integer coding technology,systematic dynamic combinations of base learners and meta-learners were generated,resulting in a total of 40 non-repetitive fusion models.The optimal combination was selected through a comprehensive evaluation based on multiple assessment indicators.[Results]The combination"PLS-DELM-MLP-LR"(code 1367)achieved coefficients of determination of 0.9732 and 0.9780 on the validation set and independent test set,respectively,with relative root mean square errors of 2.35%and 2.36%,and residual predictive deviations of 6.1075 and 6.7479,respectively.[Conclusions]The Stacking fusion model significantly enhances the predictive accuracy and robustness of spectral quantitative analysis,providing an efficient and feasible solution for modeling complex agricultural product spectral data.展开更多
This study presents a comprehensive evaluation of tropical cyclone(TC)forecast performance in the western North Pacific from 2013 to 2022,based on operational forecasts issued by the China Meteorological Administratio...This study presents a comprehensive evaluation of tropical cyclone(TC)forecast performance in the western North Pacific from 2013 to 2022,based on operational forecasts issued by the China Meteorological Administration.The analysis reveals systematic improvements in both track and intensity forecasts over the decade,with distinct error characteristics observed across various forecast parameters.Track forecast errors have steadily decreased,particularly for longer lead times,while error magnitudes have increased with longer forecast lead times.Intensity forecasts show similar progressive enhancements,with maximum sustained wind speed errors decreasing by 0.26 m/s per year for 120 h forecasts.The study also identifies several key patterns in forecast performance:typhoon-grade or stronger TCs exhibit smaller track errors than week or weaker systems;intensity forecasts systematically overestimate weaker TCs while underestimating stronger systems;and spatial error distributions show greater track inaccuracies near landmasses and regional intensity biases.These findings highlight both the significant advances in TC forecasting capability achieved through improved modeling and observational systems,and the remaining challenges in predicting TC changes and landfall behavior,providing valuable benchmarks for future forecast system development.展开更多
The solar cycle(SC),a phenomenon caused by the quasi-periodic regular activities in the Sun,occurs approximately every 11 years.Intense solar activity can disrupt the Earth’s ionosphere,affecting communication and na...The solar cycle(SC),a phenomenon caused by the quasi-periodic regular activities in the Sun,occurs approximately every 11 years.Intense solar activity can disrupt the Earth’s ionosphere,affecting communication and navigation systems.Consequently,accurately predicting the intensity of the SC holds great significance,but predicting the SC involves a long-term time series,and many existing time series forecasting methods have fallen short in terms of accuracy and efficiency.The Time-series Dense Encoder model is a deep learning solution tailored for long time series prediction.Based on a multi-layer perceptron structure,it outperforms the best previously existing models in accuracy,while being efficiently trainable on general datasets.We propose a method based on this model for SC forecasting.Using a trained model,we predict the test set from SC 19 to SC 25 with an average mean absolute percentage error of 32.02,root mean square error of 30.3,mean absolute error of 23.32,and R^(2)(coefficient of determination)of 0.76,outperforming other deep learning models in terms of accuracy and training efficiency on sunspot number datasets.Subsequently,we use it to predict the peaks of SC 25 and SC 26.For SC 25,the peak time has ended,but a stronger peak is predicted for SC 26,of 199.3,within a range of 170.8-221.9,projected to occur during April 2034.展开更多
Accurate short-term electricity price forecasts are essential for market participants to optimize bidding strategies,hedge risk and plan generation schedules.By leveraging advanced data analytics and machine learning ...Accurate short-term electricity price forecasts are essential for market participants to optimize bidding strategies,hedge risk and plan generation schedules.By leveraging advanced data analytics and machine learning methods,accurate and reliable price forecasts can be achieved.This study forecasts day-ahead prices in Türkiye’s electricity market using eXtreme Gradient Boosting(XGBoost).We benchmark XGBoost against four alternatives—Support Vector Machines(SVM),Long Short-Term Memory(LSTM),Random Forest(RF),and Gradient Boosting(GBM)—using 8760 hourly observations from 2023 provided by Energy Exchange Istanbul(EXIST).All models were trained on an identical chronological 80/20 train–test split,with hyperparameters tuned via 5-fold cross-validation on the training set.XGBoost achieved the best performance(Mean Absolute Error(MAE)=144.8 TRY/MWh,Root Mean Square Error(RMSE)=201.8 TRY/MWh,coefficient of determination(R^(2))=0.923)while training in 94 s.To enhance interpretability and identify key drivers,we employed Shapley Additive Explanations(SHAP),which highlighted a strong association between higher prices and increased natural-gas-based generation.The results provide a clear performance benchmark and practical guidance for selecting forecasting approaches in day-ahead electricity markets.展开更多
Marine heatwaves(MHWs)in the South China Sea(SCS)significantly impact marine ecosystems and socioeconomic development,yet accurately forecasting MHWs remains a challenge.This study developed an upper-ocean temperature...Marine heatwaves(MHWs)in the South China Sea(SCS)significantly impact marine ecosystems and socioeconomic development,yet accurately forecasting MHWs remains a challenge.This study developed an upper-ocean temperature forecasting model based on ConvLSTM for the northern SCS and,in conjunction with the ocean forecasting system LICOM Forecast System(LFS),constructed a hybrid Fusion model using Wasserstein-Distance optimization.The ability of these three models to forecast key MHW metrics with a 10-day lead was assessed during the summer of 2022 in the SCS.Overall,the Fusion model takes advantage of LFS and ConvLSTM,providing superior forecasts for both the duration and intensity of MHWs in the southern SCS.LFS(ConvLSTM)overestimates(underestimates)the duration of MHWs and all models exhibit limitations in forecasting the intensity of MHWs in part of the SCS.The Fusion model's superior forecast skill for MHWs may be attributable to its more realistic representation of the upper-ocean thermal structure with shallower mixed-layer depths during MHWs.This study highlights that combining the deep learning technique with a dynamical model can improve MHW forecasting and has certain physical interpretability.展开更多
With the rapid proliferation of electric vehicles,their charging loads pose new challenges to power grid stability and operational efficiency.To address this,this study employs a Monte Carlo simulation model to analyz...With the rapid proliferation of electric vehicles,their charging loads pose new challenges to power grid stability and operational efficiency.To address this,this study employs a Monte Carlo simulation model to analyze the charging load characteristics of six battery electric vehicle categories in Hebei Province,leveraging multi-source probabilistic distribution data under typical operational scenarios.The findings reveal that electric vehicle charging loads are primarily concentrated during midday and nighttime periods,with significant load fluctuations exerting substantial pressure on the grid.In response,this paper proposes strategic interventions including optimized charging infrastructure planning,time-of-use electricity pricing mechanisms,and smart charging technologies to balance grid loads.The results provide a theoretical foundation for electric vehicle load forecasting,smart grid dispatching,and vehicle-grid integration,thereby enhancing grid operational efficiency and sustainability.展开更多
This research evaluates the performance of an eddy-resolving forecast system(LFS)in simulating mesoscale eddies over the South China Sea(SCs)through a comparative analysis with satellite observations and the reanalysi...This research evaluates the performance of an eddy-resolving forecast system(LFS)in simulating mesoscale eddies over the South China Sea(SCs)through a comparative analysis with satellite observations and the reanalysis dataset from the Global Ocean Physics Reanalysis product(CMEMS).The findings indicate that the spatial characteristics of eddy kinetic energy,number,and amplitude of coherent mesoscale eddies simulated by LFS exhibit a reasonable agreement with satellite observations.The reproduced seasonal variations are also comparable to outputs from the CMEMS reanalysis dataset.Nevertheless,certain systematic biases have also been identified.In the SCS,LFS generates approximately 17%fewer eddies than observed.Such biases are also evident in the CMEMS reanalysis dataset.Similar to the statistics shown in the CMEMS reanalysis dataset,both cyclonic and anticyclonic eddies are significantly weaker in LFS compared to the observations.Additionally,the composite three-dimensional structures of mesoscale eddies simulated by LFS exhibit a remarkable similarity to those identified in the CMEMS reanalysis datasets.This work lays the foundation for further studies using LFS to investigate the predictability of mesoscale eddies and enhance the accuracy of simulations.展开更多
Deep learning-based methods have become alternatives to traditional numerical weather prediction systems,offering faster computation and the ability to utilize large historical datasets.However,the application of deep...Deep learning-based methods have become alternatives to traditional numerical weather prediction systems,offering faster computation and the ability to utilize large historical datasets.However,the application of deep learning to medium-range regional weather forecasting with limited data remains a significant challenge.In this work,three key solutions are proposed:(1)motivated by the need to improve model performance in data-scarce regional forecasting scenarios,the authors innovatively apply semantic segmentation models,to better capture spatiotemporal features and improve prediction accuracy;(2)recognizing the challenge of overfitting and the inability of traditional noise-based data augmentation methods to effectively enhance model robustness,a novel learnable Gaussian noise mechanism is introduced that allows the model to adaptively optimize perturbations for different locations,ensuring more effective learning;and(3)to address the issue of error accumulation in autoregressive prediction,as well as the challenge of learning difficulty and the lack of intermediate data utilization in one-shot prediction,the authors propose a cascade prediction approach that effectively resolves these problems while significantly improving model forecasting performance.The method achieves a competitive result in The East China Regional AI Medium Range Weather Forecasting Competition.Ablation experiments further validate the effectiveness of each component,highlighting their contributions to enhancing prediction performance.展开更多
Marine forecasting is critical for navigation safety and disaster prevention.However,traditional ocean numerical forecasting models are often limited by substantial errors and inadequate capture of temporal-spatial fe...Marine forecasting is critical for navigation safety and disaster prevention.However,traditional ocean numerical forecasting models are often limited by substantial errors and inadequate capture of temporal-spatial features.To address the limitations,the paper proposes a TimeXer-based numerical forecast correction model optimized by an exogenous-variable attention mechanism.The model treats target forecast values as internal variables,and incorporates historical temporal-spatial data and seven-day numerical forecast results from traditional models as external variables based on the embedding strategy of TimeXer.Using a self-attention structure,the model captures correlations between exogenous variables and target sequences,explores intrinsic multi-dimensional relationships,and subsequently corrects endogenous variables with the mined exogenous features.The model’s performance is evaluated using metrics including MSE(Mean Squared Error),MAE(Mean Absolute Error),RMSE(Root Mean Square Error),MAPE(Mean Absolute Percentage Error),MSPE(Mean Square Percentage Error),and computational time,with TimeXer and PatchTST models serving as benchmarks.Experiment results show that the proposed model achieves lower errors and higher correction accuracy for both one-day and seven-day forecasts.展开更多
基金Special Research Program for Public Welfare(Meteorology)of China(GYHY200906009,GYHY201006015,GYHY200906007)National Natural Science Foundation of China(4107503541475044)
文摘This study investigates multi-model ensemble forecasts of track and intensity of tropical cyclones over the western Pacific, based on forecast outputs from the China Meteorological Administration, European Centre for Medium-Range Weather Forecasts, Japan Meteorological Agency and National Centers for Environmental Prediction in the THORPEX Interactive Grand Global Ensemble(TIGGE) datasets. The multi-model ensemble schemes, namely the bias-removed ensemble mean(BREM) and superensemble(SUP), are compared with the ensemble mean(EMN) and single-model forecasts. Moreover, a new model bias estimation scheme is investigated and applied to the BREM and SUP schemes. The results showed that, compared with single-model forecasts and EMN, the multi-model ensembles of the BREM and SUP schemes can have smaller errors in most cases. However, there were also circumstances where BREM was less skillful than EMN, indicating that using a time-averaged error as model bias is not optimal. A new model bias estimation scheme of the biweight mean is introduced. Through minimizing the negative influence of singular errors, this scheme can obtain a more accurate model bias estimation and improve the BREM forecast skill. The application of the biweight mean in the bias calculation of SUP also resulted in improved skill. The results indicate that the modification of multi-model ensemble schemes through this bias estimation method is feasible.
基金supported by the National Natural Science Foundation of China(Grant Nos.42030610 and 42205006)the Startup Foundation for Introducing Talent of NUIST(2023r121)。
文摘This study focuses on an extreme rainfall event in East China during the mei-yu season,in which the capital city(Nanjing)of Jiangsu Province experienced a maximum 14-h rainfall accumulation of 209.6 mm and a peak hourly rainfall of 118.8 mm.The performance of two sets of convection-permitting ensemble forecast systems(CEFSs),each with 30 members and a 3-km horizontal grid spacing,is evaluated.The CEFS_ICBCs,using multiple initial and boundary conditions(ICs and BCs),and the CEFS_ICBCs Phys,which incorporates both multi-physics schemes and ICs/BCs,are compared to the CMA-REPS(China Meteorological Administration-Regional Ensemble Prediction System)with a coarser 10-km grid spacing.The two CEFSs demonstrate more uniform rank histograms and lower Brier scores(with higher resolution),improving precipitation intensity predictions and providing more reliable probability forecasts,although they overestimate precipitation over Mt.Dabie.It is challenging for the CEFSs to capture the evolution of mesoscale rainstorms that are known to be related to the errors in predicting the southwesterly low-level winds.Sensitivity experiments reveal that the microphysics and radiation schemes introduce considerable uncertainty in predicting the intensity and location of heavy rainfall in and near Nanjing and Mt.Dabie.In particular,the Asymmetric Convection Model 2(ACM2)planetary boundary layer scheme combined with the Pleim-Xiu surface layer scheme tends to produce a biased northeastward extension of the boundary-layer jet,contributing to the northeastward bias of heavy precipitation around Nanjing in the CEFS_ICBCs.
基金supported by the Southern Marine Science and Engineering Guangdong Laboratory(Zhuhai)(Grant No.SML2023SP214)the National Natural Science Foundation of China(Grant Nos.62071279 and 42206029)the National Key R&D Program of China(Grant No.2020YFA0608804)。
文摘Grid forecasting can be used to effectively enhance the spatial and temporal density of forecast products,thereby improving the capability of short-term marine disaster forecasting and warnings in terms of proximity.The traditional method that relies on forecasters'subjective correction of station observation data for forecasting has been unable to meet the practical needs of refined forecasting.To address this problem,this paper proposes a Transformer-enhanced UNet(TransUNet)model for wave forecast AI correction,which fuses wind and wave information.The Transformer structure is integrated into the encoder of the UNet model,and instead of using the traditional upsampling method,the dual-sampling module is employed in the decoder to enhance the feature extraction capability.This paper compares the TransUNet model with the traditional UNet model using wind speed forecast data,wave height forecast data,and significant wave height reanalysis data provided by ECMWF.The experimental results indicate that the TransUNet model yields smaller root-meansquare errors,mean errors,and standard deviations of the corrected results for the next 24-h forecasts than does the UNet model.Specifically,the root-mean-square error decreased by more than 21.55%compared to its precorrection value.According to the statistical analysis,87.81%of the corrected wave height errors for the next 24-h forecast were within±0.2m,with only 4.56%falling beyond±0.3 m.This model effectively limits the error range and enhances the ability to forecast wave heights.
文摘This study investigates the load-bearing capacity of open-ended pipe piles in sandy soil, with a specific focus on the impact of soil plug constraints at four levels(no plug, 25% plug, 50% plug, and full plug). Leveraging a dataset comprising open-ended pipe piles with varying geometrical and geotechnical properties, this research employs shallow neural network(SNN) and deep neural network(DNN) models to predict plugging conditions for both driven and pressed installation types. This paper underscores the importance of key parameters such as the settlement value,applied load, installation type, and soil configuration(loose, medium, and dense) in accurately predicting pile settlement. These findings offer valuable insights for optimizing pile design and construction in geotechnical engineering,addressing a longstanding challenge in the field. The study demonstrates the potential of the SNN and DNN models in precisely identifying plugging conditions before pile driving, with the SNN achieving R2 values ranging from0.444 to 0.711 and RMSPE values ranging from 24.621% to 48.663%, whereas the DNN exhibits superior performance, with R2 values ranging from 0.815 to 0.942 and RMSPE values ranging from 4.419% to 10.325%. These results have significant implications for enhancing construction practices and reducing uncertainties associated with pile foundation projects in addition to leveraging artificial intelligence tools to avoid long experimental procedures.
基金supported by the Strategic Research and Consulting Project of the Chinese Academy of Engineering[grant number 2024-XBZD-14]the National Natural Science Foundation of China[grant numbers 42192553 and 41922036]the Fundamental Research Funds for the Central Universities–Cemac“GeoX”Interdisciplinary Program[grant number 020714380207]。
文摘The impacts of lateral boundary conditions(LBCs)provided by numerical models and data-driven networks on convective-scale ensemble forecasts are investigated in this study.Four experiments are conducted on the Hangzhou RDP(19th Hangzhou Asian Games Research Development Project on Convective-scale Ensemble Prediction and Application)testbed,with the LBCs respectively sourced from National Centers for Environmental Prediction(NCEP)Global Forecast System(GFS)forecasts with 33 vertical levels(Exp_GFS),Pangu forecasts with 13 vertical levels(Exp_Pangu),Fuxi forecasts with 13 vertical levels(Exp_Fuxi),and NCEP GFS forecasts with the vertical levels reduced to 13(the same as those of Exp_Pangu and Exp_Fuxi)(Exp_GFSRDV).In general,Exp_Pangu performs comparably to Exp_GFS,while Exp_Fuxi shows slightly inferior performance compared to Exp_Pangu,possibly due to its less accurate large-scale predictions.Therefore,the ability of using data-driven networks to efficiently provide LBCs for convective-scale ensemble forecasts has been demonstrated.Moreover,Exp_GFSRDV has the worst convective-scale forecasts among the four experiments,which indicates the potential improvement of using data-driven networks for LBCs by increasing the vertical levels of the networks.However,the ensemble spread of the four experiments barely increases with lead time.Thus,each experiment has insufficient ensemble spread to present realistic forecast uncertainties,which will be investigated in a future study.
基金supported by the European Commission Horizon 2020 Framework Program No.861584the Taishan Distinguished Professor Fund No.20190910.
文摘The financial health of leading enterprises has a significant impact on the sustainable development of the global economy.Most data-driven financial health forecasts are based on the direct use of small-scale machine learning.In this study,we proposed the idea of optimization coupling learning to improve these machine learning models in financial health forecasting.It not only revealed lagging,immediate,continuous impacts of various indicators in different fiscal year,but also had the same low computational cost and complexity as known small-scale machine learning models.We used our optimization coupling learning to investigate 3424 leading enterprises in China and revealed inner triggering mechanisms and differences of enterprises’financial health status from individual behavior to macro level.
基金supported by the National Natural Science Foundation of China(Grant No.42192553,Grant No.41805070)Open Grants of the State Key Laboratory of Severe Weather(2024LASW-B05)+7 种基金Natural Science Fund of Anhui Province of China under grant(2308085MD127)the China Meteorological Administration Tornado Key Laboratory(TKL202306)Beijige Funding from Jiangsu Research Institute of Meteorological Science(BJG202503)the Open Grants of China Meteorological Administration Radar Meteorology Key Laboratory(2023LRM-B03)the Open Project Fund of China Meteorological Administration Basin Heavy Rainfall Key Laboratory(2023BHRY20)the Shanghai Typhoon Research Foundation(TFJJ202107)Innovation and Development Projects of Anhui Provincial Meteorological Bureau(CXM202205)the High Performance Computing Center of Nanjing University of Information Science&Technology for their support of this work.
文摘Moisture conditions are crucial for the maintenance and development of severe convection.In the indirect assimilation of radar reflectivity,hydrometeors and water vapor retrieved from reflectivity are assimilated to avoid the nonlinearity issues associated with the observation operator.In a widely applied water vapor retrieval scheme,a cloud is assumed to be saturated when the radar reflectivity exceeds a certain threshold.This study replaces the traditional retrieval scheme with the“Z-RH”(radar reflectivity and relative humidity)linear statistical relationship for estimating the water vapor content,which is implemented to reduce the uncertainty caused by empirical relationships.The“Z-RH”relationship is statistically obtained from the humidity and the observations for rainfall rate at different temperature intervals with the use of the Z-R(radar reflectivity-rain rate)relationship.The impacts of these two retrieval approaches are investigated in the analyses and forecasts based on the radar reflectivity.The results suggest that both water vapor retrieval schemes yield similar reflectivity analyses,with“Z-RH”showing slightly stronger reflectivity intensities.Utilizing a“Z-RH”scheme contributes significantly to the improved analyses and forecasts of humidity and wind fields,resulting in more reasonable thermodynamic and dynamic structures.As the“Z-RH”relationship obtained by real-time statistics in a specific area provides a scientific basis for the retrieval of water vapor,a“Z-RH”scheme is beneficial to obtain more accurate reflectivity forecasts.The overall scores for the predicted precipitation of a“Z-RH”scheme are roughly 10%-20%higher compared to those of the traditional scheme.
基金supported by the National Natural Science Foundation of China (No.U2039207)
文摘Symposium overview The International Symposium on Earthquake Forecasting to Commemorate the 50th Anniversary of the 1975 Haicheng M7.3 Earthquake,Liaoning,China,was held in Shenyang,China,from 8 to 11 July 2025.The symposium was organized by the Institute of Earthquake Forecasting,China Earthquake Administration(CEA),the State Key Laboratory of Earthquake Dynamics and Forecasting,and the China Seismic Experimental Site(CSES),in collaboration with the International Association of Seismology and Physics of the Earth’s Interior(IASPEI),the APEC Cooperation for Earthquake Science(ACES).
基金Supported by Traditional Chinese Medicine Standardization Project of National Administration of Traditional Chinese Medicine:Research on the Physical Characteristics and Pre-disease Health Management of the Elderly in Hubei Province(No.GZY-FJS-2022-046)。
文摘OBJECTIVE:To develop an automated system for identifying and classifying constitution types in Traditional Chinese Medicine(TCM)by leveraging multi-model fusion algorithms.METHODS:A condensed version of a physical information collection form was designed to facilitate efficient data acquisition.The collected data were analyzed using a multi-model fusion approach,which integrated several machine learning techniques.These included support vector machines,Naive Bayes,decision trees,random forests,logistic regression,multilayer perceptrons,K-nearest neighbors,gradient boosting,adaptive ensemble learning,and recurrent neural networks.A soft voting strategy was used to combine the predictive outputs of each model,enabling the selection of the most effective model combination.RESULTS:The classification models demonstrated consistent and robust performance across most TCM constitution types when enhanced by the multi-model fusion strategy.In particular,high levels of accuracy,precision,recall,and F1-score were achieved for constitution types such as Yang deficiency,Qi deficiency,and Qi stagnation.However,the classification performance for the Yin deficiency constitution was relatively lower,indicating the need for further refinement and optimization in future research.CONCLUSION:This study introduces a novel,automated method for classifying TCM constitution types through the application of multi-model fusion algorithms.The approach simplifies the complex task of constitution identification while offering a practical and theoretical framework for the intelligent diagnosis of TCM body types.The findings have the potential to enhance personalized health management and support clinical decision-making in TCM diagnosis and treatment.
基金Science and Technology Development Program of the“Taihu Light”(K20231023)CMA Numerical Weather Prediction R&D Project(TCYF2024QH007)+1 种基金“Qing Lan”Project of Jiangsu Province for C.H.LUWuxi University Research Start-up Fund for Introduced Talents(2023r037)。
文摘Since the initiation of the subseasonal-to-seasonal prediction project by the World Meteorological Organization,the accuracy of model forecasts has improved notably.However,substantial discrepancies have been observed among forecast results produced by different ensemble members when applied to South China.To enhance the accuracy of sub-seasonal forecasts in this region,it is essential to develop new methods that can effectively leverage multiple predictive models.This study introduces a weighted ensemble forecasting method based on online learning to improve forecast accuracy.We utilized ensemble forecasts from three models:the Integrated Forecasting System model from the European Centre for Medium-Range Weather Forecasts,the Climate Forecast System Version 2 model from the National Centers for Environmental Prediction,and the Beijing Climate Center-Climate Prediction System version 3 model from the China Meteorological Administration.The ensemble weights are trained using an online learning approach.The results indicate that the forecasts obtained through online learning outperform those of the original dynamical models.Compared to the simple ensemble results of the three models,the weighted ensemble model showed a stronger capability to capture temperature and precipitation patterns in South China.Therefore,this method has the potential to improve the accuracy of sub-seasonal forecasts in this region.
基金supported by the National Natural Science Foundation of China(No.52171284)。
文摘Wind speed is a crucial parameter affecting wind energy utilization.However,its volatility leads to time-varying power output.Herein,a novel Seq2Seq model integrating deep learning,data denoising,and a shape-aware loss function is proposed for accurate multistep wind speed forecasting.In this model,the wind speed data is first denoised using the maximal overlap discrete wavelet transform.Next,an encoder-decoder network based on a temporal convolutional network,bidirectional gated recurrent unit,and multihead self-attention is employed for forecasting.Additionally,to enhance the ability of the model to identify temporal dynamics,a shape-aware loss function,ITILDE-Q,is employed in the model.To verify the effectiveness of the proposed model,a comparative experiment and an ablation experiment were conducted using three datasets of measured wind speeds.Three error metrics and a similarity metric were adopted for comprehensive evaluation.The experimental results showed that the proposed model consistently outperforms benchmark models in all tested forecasting scenarios,with particularly pronounced differences in performance over longer forecast horizons.Furthermore,the synergistic interaction of the three key components contributes to the extraordinary performance of the proposed model.
文摘[Objectives]This study was conducted to achieve rapid and accurate detection of protein content in rice with a particle size of 1.0 mm.[Methods]A multi-model fusion strategy was proposed on the basis of Stacking ensemble learning.A base learner pool was constructed,containing Partial Least Squares(PLS),Support Vector Machine(SVM),Deep Extreme Learning Machine(DELM),Random Forest(RF),Gradient Boosting Decision Tree(GBDT),and Multilayer Perceptron(MLP).PLS,DELM,and Linear Regression(LR)were used as meta-learner candidates.Employing integer coding technology,systematic dynamic combinations of base learners and meta-learners were generated,resulting in a total of 40 non-repetitive fusion models.The optimal combination was selected through a comprehensive evaluation based on multiple assessment indicators.[Results]The combination"PLS-DELM-MLP-LR"(code 1367)achieved coefficients of determination of 0.9732 and 0.9780 on the validation set and independent test set,respectively,with relative root mean square errors of 2.35%and 2.36%,and residual predictive deviations of 6.1075 and 6.7479,respectively.[Conclusions]The Stacking fusion model significantly enhances the predictive accuracy and robustness of spectral quantitative analysis,providing an efficient and feasible solution for modeling complex agricultural product spectral data.
基金supported by the National Key R&D Program of China [grant number 2023YFC3008004]。
文摘This study presents a comprehensive evaluation of tropical cyclone(TC)forecast performance in the western North Pacific from 2013 to 2022,based on operational forecasts issued by the China Meteorological Administration.The analysis reveals systematic improvements in both track and intensity forecasts over the decade,with distinct error characteristics observed across various forecast parameters.Track forecast errors have steadily decreased,particularly for longer lead times,while error magnitudes have increased with longer forecast lead times.Intensity forecasts show similar progressive enhancements,with maximum sustained wind speed errors decreasing by 0.26 m/s per year for 120 h forecasts.The study also identifies several key patterns in forecast performance:typhoon-grade or stronger TCs exhibit smaller track errors than week or weaker systems;intensity forecasts systematically overestimate weaker TCs while underestimating stronger systems;and spatial error distributions show greater track inaccuracies near landmasses and regional intensity biases.These findings highlight both the significant advances in TC forecasting capability achieved through improved modeling and observational systems,and the remaining challenges in predicting TC changes and landfall behavior,providing valuable benchmarks for future forecast system development.
基金supported by the Academic Research Projects of Beijing Union University(ZK20202204)the National Natural Science Foundation of China(12250005,12073040,12273059,11973056,12003051,11573037,12073041,11427901,11572005,11611530679 and 12473052)+1 种基金the Strategic Priority Research Program of the China Academy of Sciences(XDB0560000,XDA15052200,XDB09040200,XDA15010700,XDB0560301,and XDA15320102)the Chinese Meridian Project(CMP).
文摘The solar cycle(SC),a phenomenon caused by the quasi-periodic regular activities in the Sun,occurs approximately every 11 years.Intense solar activity can disrupt the Earth’s ionosphere,affecting communication and navigation systems.Consequently,accurately predicting the intensity of the SC holds great significance,but predicting the SC involves a long-term time series,and many existing time series forecasting methods have fallen short in terms of accuracy and efficiency.The Time-series Dense Encoder model is a deep learning solution tailored for long time series prediction.Based on a multi-layer perceptron structure,it outperforms the best previously existing models in accuracy,while being efficiently trainable on general datasets.We propose a method based on this model for SC forecasting.Using a trained model,we predict the test set from SC 19 to SC 25 with an average mean absolute percentage error of 32.02,root mean square error of 30.3,mean absolute error of 23.32,and R^(2)(coefficient of determination)of 0.76,outperforming other deep learning models in terms of accuracy and training efficiency on sunspot number datasets.Subsequently,we use it to predict the peaks of SC 25 and SC 26.For SC 25,the peak time has ended,but a stronger peak is predicted for SC 26,of 199.3,within a range of 170.8-221.9,projected to occur during April 2034.
文摘Accurate short-term electricity price forecasts are essential for market participants to optimize bidding strategies,hedge risk and plan generation schedules.By leveraging advanced data analytics and machine learning methods,accurate and reliable price forecasts can be achieved.This study forecasts day-ahead prices in Türkiye’s electricity market using eXtreme Gradient Boosting(XGBoost).We benchmark XGBoost against four alternatives—Support Vector Machines(SVM),Long Short-Term Memory(LSTM),Random Forest(RF),and Gradient Boosting(GBM)—using 8760 hourly observations from 2023 provided by Energy Exchange Istanbul(EXIST).All models were trained on an identical chronological 80/20 train–test split,with hyperparameters tuned via 5-fold cross-validation on the training set.XGBoost achieved the best performance(Mean Absolute Error(MAE)=144.8 TRY/MWh,Root Mean Square Error(RMSE)=201.8 TRY/MWh,coefficient of determination(R^(2))=0.923)while training in 94 s.To enhance interpretability and identify key drivers,we employed Shapley Additive Explanations(SHAP),which highlighted a strong association between higher prices and increased natural-gas-based generation.The results provide a clear performance benchmark and practical guidance for selecting forecasting approaches in day-ahead electricity markets.
基金supported by the National Natural Science Foundation of China [grant numbers 42375168 and 42205035]a Shanghai Science and Technology Commission Project [grant number 23DZ1204704]。
文摘Marine heatwaves(MHWs)in the South China Sea(SCS)significantly impact marine ecosystems and socioeconomic development,yet accurately forecasting MHWs remains a challenge.This study developed an upper-ocean temperature forecasting model based on ConvLSTM for the northern SCS and,in conjunction with the ocean forecasting system LICOM Forecast System(LFS),constructed a hybrid Fusion model using Wasserstein-Distance optimization.The ability of these three models to forecast key MHW metrics with a 10-day lead was assessed during the summer of 2022 in the SCS.Overall,the Fusion model takes advantage of LFS and ConvLSTM,providing superior forecasts for both the duration and intensity of MHWs in the southern SCS.LFS(ConvLSTM)overestimates(underestimates)the duration of MHWs and all models exhibit limitations in forecasting the intensity of MHWs in part of the SCS.The Fusion model's superior forecast skill for MHWs may be attributable to its more realistic representation of the upper-ocean thermal structure with shallower mixed-layer depths during MHWs.This study highlights that combining the deep learning technique with a dynamical model can improve MHW forecasting and has certain physical interpretability.
基金funded by Humanities and Social Sciences of Ministry of Education Planning Fund of China,grant number 21YJA790009National Natural Science Foundation of China,grant number 72140001.
文摘With the rapid proliferation of electric vehicles,their charging loads pose new challenges to power grid stability and operational efficiency.To address this,this study employs a Monte Carlo simulation model to analyze the charging load characteristics of six battery electric vehicle categories in Hebei Province,leveraging multi-source probabilistic distribution data under typical operational scenarios.The findings reveal that electric vehicle charging loads are primarily concentrated during midday and nighttime periods,with significant load fluctuations exerting substantial pressure on the grid.In response,this paper proposes strategic interventions including optimized charging infrastructure planning,time-of-use electricity pricing mechanisms,and smart charging technologies to balance grid loads.The results provide a theoretical foundation for electric vehicle load forecasting,smart grid dispatching,and vehicle-grid integration,thereby enhancing grid operational efficiency and sustainability.
基金supported by the National Key R&D Program for Developing Basic Sciences [grant number 2022YFC3104805]the National Natural Science Foundation of China [grant numbers 92358302 and 42306219]+1 种基金supported by the Tai Shan Scholar Program [grant number tstp20231237]Laoshan Laboratory project [grant number LSKJ202300301]。
文摘This research evaluates the performance of an eddy-resolving forecast system(LFS)in simulating mesoscale eddies over the South China Sea(SCs)through a comparative analysis with satellite observations and the reanalysis dataset from the Global Ocean Physics Reanalysis product(CMEMS).The findings indicate that the spatial characteristics of eddy kinetic energy,number,and amplitude of coherent mesoscale eddies simulated by LFS exhibit a reasonable agreement with satellite observations.The reproduced seasonal variations are also comparable to outputs from the CMEMS reanalysis dataset.Nevertheless,certain systematic biases have also been identified.In the SCS,LFS generates approximately 17%fewer eddies than observed.Such biases are also evident in the CMEMS reanalysis dataset.Similar to the statistics shown in the CMEMS reanalysis dataset,both cyclonic and anticyclonic eddies are significantly weaker in LFS compared to the observations.Additionally,the composite three-dimensional structures of mesoscale eddies simulated by LFS exhibit a remarkable similarity to those identified in the CMEMS reanalysis datasets.This work lays the foundation for further studies using LFS to investigate the predictability of mesoscale eddies and enhance the accuracy of simulations.
基金supported by the National Natural Science Foundation of China[grant number 62376217]the Young Elite Scientists Sponsorship Program by CAST[grant number 2023QNRC001]the Joint Research Project for Meteorological Capacity Improvement[grant number 24NLTSZ003]。
文摘Deep learning-based methods have become alternatives to traditional numerical weather prediction systems,offering faster computation and the ability to utilize large historical datasets.However,the application of deep learning to medium-range regional weather forecasting with limited data remains a significant challenge.In this work,three key solutions are proposed:(1)motivated by the need to improve model performance in data-scarce regional forecasting scenarios,the authors innovatively apply semantic segmentation models,to better capture spatiotemporal features and improve prediction accuracy;(2)recognizing the challenge of overfitting and the inability of traditional noise-based data augmentation methods to effectively enhance model robustness,a novel learnable Gaussian noise mechanism is introduced that allows the model to adaptively optimize perturbations for different locations,ensuring more effective learning;and(3)to address the issue of error accumulation in autoregressive prediction,as well as the challenge of learning difficulty and the lack of intermediate data utilization in one-shot prediction,the authors propose a cascade prediction approach that effectively resolves these problems while significantly improving model forecasting performance.The method achieves a competitive result in The East China Regional AI Medium Range Weather Forecasting Competition.Ablation experiments further validate the effectiveness of each component,highlighting their contributions to enhancing prediction performance.
基金supported by the National Key Research and Development Program Project(2023YFC3107804)Planning Fund Project of Humanities and Social Sciences Research of the Ministry of Education(24YJA880097)the Graduate Education Reform Project in North China University of Technology(217051360025XN095-17)。
文摘Marine forecasting is critical for navigation safety and disaster prevention.However,traditional ocean numerical forecasting models are often limited by substantial errors and inadequate capture of temporal-spatial features.To address the limitations,the paper proposes a TimeXer-based numerical forecast correction model optimized by an exogenous-variable attention mechanism.The model treats target forecast values as internal variables,and incorporates historical temporal-spatial data and seven-day numerical forecast results from traditional models as external variables based on the embedding strategy of TimeXer.Using a self-attention structure,the model captures correlations between exogenous variables and target sequences,explores intrinsic multi-dimensional relationships,and subsequently corrects endogenous variables with the mined exogenous features.The model’s performance is evaluated using metrics including MSE(Mean Squared Error),MAE(Mean Absolute Error),RMSE(Root Mean Square Error),MAPE(Mean Absolute Percentage Error),MSPE(Mean Square Percentage Error),and computational time,with TimeXer and PatchTST models serving as benchmarks.Experiment results show that the proposed model achieves lower errors and higher correction accuracy for both one-day and seven-day forecasts.
