Predicting player performance in sports is a critical challenge with significant implications for team success,fan engagement,and financial outcomes.Although,inMajor League Baseball(MLB),statistical methodologies such...Predicting player performance in sports is a critical challenge with significant implications for team success,fan engagement,and financial outcomes.Although,inMajor League Baseball(MLB),statistical methodologies such as sabermetrics have been widely used,the dynamic nature of sports makes accurate performance prediction a difficult task.Enhanced forecasts can provide immense value to team managers by aiding strategic player contract and acquisition decisions.This study addresses this challenge by employing the temporal fusion transformer(TFT),an advanced and cutting-edge deep learning model for complex data,to predict pitchers’earned run average(ERA),a key metric in baseball performance analysis.The performance of the TFT model is evaluated against recurrent neural network-based approaches and existing projection systems.In experimental results,the TFT based model consistently outperformed its counterparts,demonstrating superior accuracy in pitcher performance prediction.By leveraging the advanced capabilities of TFT,this study contributes to more precise player evaluations and improves strategic planning in baseball.展开更多
Surface electromyography(sEMG)-based gesture recognition is a key technology in the field of human–computer interaction.However,existing gesture recognition methods face challenges in effectively integrating discrimi...Surface electromyography(sEMG)-based gesture recognition is a key technology in the field of human–computer interaction.However,existing gesture recognition methods face challenges in effectively integrating discriminative temporal feature representations from sEMG signals.In this paper,we propose a deep learning framework named TFN-FICFM comprises a Temporal Fusion Network(TFN)and Fuzzy Integral-Based Classifier Fusion method(FICFM)to improve the accuracy and robustness of gesture recognition.Firstly,we design a TFN module,which utilizes an attention-based recurrent multi-scale convolutional module to acquire multi-level temporal feature representations and achieves deep fusion of temporal features through a feature pyramid module.Secondly,the deep-fused temporal features are utilized to generate multiple sets of gesture category prediction confidences through a feedback loop.Finally,we employ FICFM to perform fuzzy fusion on prediction confidences,resulting in the ultimate decision.This study conducts extensive comparisons and ablation studies using the publicly available datasets Ninapro DB2 and DB5.Results demonstrate that the TFN-FICFM model outperforms state-of-the-art methods in classification performance.This research can serve as a benchmark for sEMG-based gesture recognition and related deep learning modeling.展开更多
Accurate estimation of land surface solar irradiation is critical for effective solar energy utilization and planning of solar photovoltaic planning.Although traditional machine learning methods have been demonstrated...Accurate estimation of land surface solar irradiation is critical for effective solar energy utilization and planning of solar photovoltaic planning.Although traditional machine learning methods have been demonstrated to estimate solar irradiation effectively,they face challenges in modeling over large regions,as well as lacking of ability to model spatial diversity and temporal dynamics of solar irradiation,and providing limited interpretability.To address these limitations,this study proposed a geospatial artificial intelligence framework augmented by Temporal Fusion Transformer for hourly estimation of land surface solar irradiation.As a case study in Australia,the results demonstrate superior performance with the coefficient of the determination,the mean absolute error,and Root Mean Square Error as high as 0.90,0.25(kWh/m^(2)),and 0.63(kWh/m^(2)),showing improvements of 21.62–66.67%,78.37–85.98%,and 62.81–73.25%,respectively,compared to the benchmarks of other methods,including Support Vector Regression,Random Forest,Gradient Boosting Machine,AdaBoost,Long Short-Term Memory,Temporal Convolutional Network,ConvLSTM,Transformer,and Graph Neural Network.Furthermore,interpretability results of the model indicate that among the temporal variables,observed solar irradiation and clear sky solar irradiation significantly contribute to the model’s performance.The results show this framework enhanced accuracy and interpretability for solar irradiation estimation over large areas,providing valuable insights for future studies and supporting decision-making for developing the renewable energy industry.展开更多
Building integrated energy systems(BIESs)are pivotal for enhancing energy efficiency by accounting for a significant proportion of global energy consumption.Two key barriers that reduce the BIES operational efficiency...Building integrated energy systems(BIESs)are pivotal for enhancing energy efficiency by accounting for a significant proportion of global energy consumption.Two key barriers that reduce the BIES operational efficiency mainly lie in the renewable generation uncertainty and operational non-convexity of combined heat and power(CHP)units.To this end,this paper proposes a soft actor-critic(SAC)algorithm to solve the scheduling problem of BIES,which overcomes the model non-convexity and shows advantages in robustness and generalization.This paper also adopts a temporal fusion transformer(TFT)to enhance the optimal solution for the SAC algorithm by forecasting the renewable generation and energy demand.The TFT can effectively capture the complex temporal patterns and dependencies that span multiple steps.Furthermore,its forecasting results are interpretable due to the employment of a self-attention layer so as to assist in more trustworthy decision-making in the SAC algorithm.The proposed hybrid data-driven approach integrating TFT and SAC algorithm,i.e.,TFT-SAC approach,is trained and tested on a real-world dataset to validate its superior performance in reducing the energy cost and computational time compared with the benchmark approaches.The generalization performance for the scheduling policy,as well as the sensitivity analysis,are examined in the case studies.展开更多
Fine spatial and temporal resolution land surface temperature(LST)data are of great importance for various researches and applications.Spatio-temporal fusion provides an important solution to obtain fine spatio-tempor...Fine spatial and temporal resolution land surface temperature(LST)data are of great importance for various researches and applications.Spatio-temporal fusion provides an important solution to obtain fine spatio-temporal resolution LST.For example,100-m,daily LST data can be created by fusing 1-km,daily Moderate Resolution Imaging Spectroradiometer(MODIS)LST with 100-m,16-day Landsat LST data.However,the quality of MODIS LST products has been decreasing noticeably in recent years,which has a great impact on fusion accuracy.To address this issue,this paper proposes to use Visible Infrared Imaging Radiometer Suite(VIIRS)LST to replace MODIS LST in spatio-temporal fusion.Meanwhile,to cope with the data discrepancy caused by the large difference in overpass time between VIIRS LST and Landsat LST,a spatio-temporal fusion method based on the Restormer(RES-STF)is proposed.Specifically,to effectively model the differences between the 2 types of data,RES-STF uses Transformer modules in Restormer,which combines the advantages of convolutional neural networks(CNN)and Transformer to effectively capture both local and global context in images.In addition,the calculation of self-attention is re-designed by concatenating CNN to increase the efficiency of feature extraction.Experimental results on 3 areas validated the effectiveness of RES-STF,which outperforms one non-deep learning-and 3 deep learning-based spatio-temporal fusion methods.Moreover,compared to MODIS LST,VIIRS LST data contain richer spatial texture information,leading to more accurate fusion results,with both RMSE and MAE reduced by about 0.5 K.展开更多
Remote sensing Change Detection(CD)involves identifying changing regions of interest in bi-temporal remote sensing images.CD technology has rapidly developed in recent years through the powerful learning ability of Co...Remote sensing Change Detection(CD)involves identifying changing regions of interest in bi-temporal remote sensing images.CD technology has rapidly developed in recent years through the powerful learning ability of Convolutional Neural Networks(CNN),affording complex feature extraction.However,the local receptive fields in the CNN limit modeling long-range contextual relationships in semantic changes.Therefore,this work explores the great potential of Siamese Transformers in CD tasks and proposes a general CD model entitled STCD that relies on Swin Transformers.In the encoding process,pure Transformers without CNN are used to model the long-range context of semantic tokens,reducing computational overhead and improving model efficiency compared to current methods.During the decoding process,the 3D convolution block obtains the changing features in the time series and generates the predicted change map in the deconvolution layer with axial attention.Extensive experiments on three binary CD datasets and one semantic CD dataset demonstrate that the proposed STCD model outperforms several popular benchmark methods considering performance and the required parameters.Among the STCD variants,the F1-Score of the Base-STCD on the three binary CD datasets LEVIR,DSIFN,and SVCD reached 89.85%,54.72%,and 93.75%,respectively,and the mF1-Score and mIoU on the semantic CD dataset SECOND were 75.60%and 66.19%.展开更多
Increased penetration of renewables for power generation has negatively impacted the dynamics of conventional fossil fuel-based power plants.The power plants operating on the base load are forced to cycle,to adjust to...Increased penetration of renewables for power generation has negatively impacted the dynamics of conventional fossil fuel-based power plants.The power plants operating on the base load are forced to cycle,to adjust to the fluctuating power demands.This results in an inefficient operation of the coal power plants,which leads up to higher operating losses.To overcome such operational challenge associated with cycling and to develop an optimal process control,this work analyzes a set of models for predicting power generation.Moreover,the power generation is intrinsically affected by the state of the power plant components,and therefore our model development also incorporates additional power plant process variables while forecasting the power generation.We present and compare multiple state-of-the-art forecasting data-driven methods for power generation to determine the most adequate and accurate model.We also develop an interpretable attention-based transformer model to explain the importance of process variables during training and forecasting.The trained deep neural network(DNN)LSTM model has good accuracy in predicting gross power generation under various prediction horizons with/without cycling events and outperforms the other models for long-term forecasting.The DNN memory-based models show significant superiority over other state-of-the-art machine learning models for short,medium and long range predictions.The transformer-based model with attention enhances the selection of historical data for multi-horizon forecasting,and also allows to interpret the significance of internal power plant components on the power generation.This newly gained insights can be used by operation engineers to anticipate and monitor the health of power plant equipment during high cycling periods.展开更多
基金supported by SKKU Global Research Platform Research Fund,Sungkyunkwan University,2024-2025.
文摘Predicting player performance in sports is a critical challenge with significant implications for team success,fan engagement,and financial outcomes.Although,inMajor League Baseball(MLB),statistical methodologies such as sabermetrics have been widely used,the dynamic nature of sports makes accurate performance prediction a difficult task.Enhanced forecasts can provide immense value to team managers by aiding strategic player contract and acquisition decisions.This study addresses this challenge by employing the temporal fusion transformer(TFT),an advanced and cutting-edge deep learning model for complex data,to predict pitchers’earned run average(ERA),a key metric in baseball performance analysis.The performance of the TFT model is evaluated against recurrent neural network-based approaches and existing projection systems.In experimental results,the TFT based model consistently outperformed its counterparts,demonstrating superior accuracy in pitcher performance prediction.By leveraging the advanced capabilities of TFT,this study contributes to more precise player evaluations and improves strategic planning in baseball.
基金supported in part by the Natural Science Foundation of Zhejiang Province(LQ23F030015)the Key Laboratory of Intelligent Processing Technology for Digital Music(Zhejiang Conservatory of Music),Ministry of Culture and Tourism(2023DMKLC013).
文摘Surface electromyography(sEMG)-based gesture recognition is a key technology in the field of human–computer interaction.However,existing gesture recognition methods face challenges in effectively integrating discriminative temporal feature representations from sEMG signals.In this paper,we propose a deep learning framework named TFN-FICFM comprises a Temporal Fusion Network(TFN)and Fuzzy Integral-Based Classifier Fusion method(FICFM)to improve the accuracy and robustness of gesture recognition.Firstly,we design a TFN module,which utilizes an attention-based recurrent multi-scale convolutional module to acquire multi-level temporal feature representations and achieves deep fusion of temporal features through a feature pyramid module.Secondly,the deep-fused temporal features are utilized to generate multiple sets of gesture category prediction confidences through a feedback loop.Finally,we employ FICFM to perform fuzzy fusion on prediction confidences,resulting in the ultimate decision.This study conducts extensive comparisons and ablation studies using the publicly available datasets Ninapro DB2 and DB5.Results demonstrate that the TFN-FICFM model outperforms state-of-the-art methods in classification performance.This research can serve as a benchmark for sEMG-based gesture recognition and related deep learning modeling.
基金substantially funded by the General Research Fund(Grant No.15603923 and 15603920)the Collaborative Research Fund(Grant No.C5062-21GF)+1 种基金Young Collaborate Research Fund(Grant No.C6003-22Y)from the Research Grants Council,Hong Kong,Chinafunding support(Grant No.BBG2 and CD81)from the Research Institute for Sustainable Urban Develop-ment,Research Institute of Land and Space,The Hong Kong Polytechnic University,Kowloon,Hong Kong,China.
文摘Accurate estimation of land surface solar irradiation is critical for effective solar energy utilization and planning of solar photovoltaic planning.Although traditional machine learning methods have been demonstrated to estimate solar irradiation effectively,they face challenges in modeling over large regions,as well as lacking of ability to model spatial diversity and temporal dynamics of solar irradiation,and providing limited interpretability.To address these limitations,this study proposed a geospatial artificial intelligence framework augmented by Temporal Fusion Transformer for hourly estimation of land surface solar irradiation.As a case study in Australia,the results demonstrate superior performance with the coefficient of the determination,the mean absolute error,and Root Mean Square Error as high as 0.90,0.25(kWh/m^(2)),and 0.63(kWh/m^(2)),showing improvements of 21.62–66.67%,78.37–85.98%,and 62.81–73.25%,respectively,compared to the benchmarks of other methods,including Support Vector Regression,Random Forest,Gradient Boosting Machine,AdaBoost,Long Short-Term Memory,Temporal Convolutional Network,ConvLSTM,Transformer,and Graph Neural Network.Furthermore,interpretability results of the model indicate that among the temporal variables,observed solar irradiation and clear sky solar irradiation significantly contribute to the model’s performance.The results show this framework enhanced accuracy and interpretability for solar irradiation estimation over large areas,providing valuable insights for future studies and supporting decision-making for developing the renewable energy industry.
文摘Building integrated energy systems(BIESs)are pivotal for enhancing energy efficiency by accounting for a significant proportion of global energy consumption.Two key barriers that reduce the BIES operational efficiency mainly lie in the renewable generation uncertainty and operational non-convexity of combined heat and power(CHP)units.To this end,this paper proposes a soft actor-critic(SAC)algorithm to solve the scheduling problem of BIES,which overcomes the model non-convexity and shows advantages in robustness and generalization.This paper also adopts a temporal fusion transformer(TFT)to enhance the optimal solution for the SAC algorithm by forecasting the renewable generation and energy demand.The TFT can effectively capture the complex temporal patterns and dependencies that span multiple steps.Furthermore,its forecasting results are interpretable due to the employment of a self-attention layer so as to assist in more trustworthy decision-making in the SAC algorithm.The proposed hybrid data-driven approach integrating TFT and SAC algorithm,i.e.,TFT-SAC approach,is trained and tested on a real-world dataset to validate its superior performance in reducing the energy cost and computational time compared with the benchmark approaches.The generalization performance for the scheduling policy,as well as the sensitivity analysis,are examined in the case studies.
基金supported by the National Natural Science Foundation of China under grants 42171345 and 42222108.
文摘Fine spatial and temporal resolution land surface temperature(LST)data are of great importance for various researches and applications.Spatio-temporal fusion provides an important solution to obtain fine spatio-temporal resolution LST.For example,100-m,daily LST data can be created by fusing 1-km,daily Moderate Resolution Imaging Spectroradiometer(MODIS)LST with 100-m,16-day Landsat LST data.However,the quality of MODIS LST products has been decreasing noticeably in recent years,which has a great impact on fusion accuracy.To address this issue,this paper proposes to use Visible Infrared Imaging Radiometer Suite(VIIRS)LST to replace MODIS LST in spatio-temporal fusion.Meanwhile,to cope with the data discrepancy caused by the large difference in overpass time between VIIRS LST and Landsat LST,a spatio-temporal fusion method based on the Restormer(RES-STF)is proposed.Specifically,to effectively model the differences between the 2 types of data,RES-STF uses Transformer modules in Restormer,which combines the advantages of convolutional neural networks(CNN)and Transformer to effectively capture both local and global context in images.In addition,the calculation of self-attention is re-designed by concatenating CNN to increase the efficiency of feature extraction.Experimental results on 3 areas validated the effectiveness of RES-STF,which outperforms one non-deep learning-and 3 deep learning-based spatio-temporal fusion methods.Moreover,compared to MODIS LST,VIIRS LST data contain richer spatial texture information,leading to more accurate fusion results,with both RMSE and MAE reduced by about 0.5 K.
基金supported by the Military Commission Science and Technology Committee Leading Fund[grant number 18-163-00-TS-004-080-01].
文摘Remote sensing Change Detection(CD)involves identifying changing regions of interest in bi-temporal remote sensing images.CD technology has rapidly developed in recent years through the powerful learning ability of Convolutional Neural Networks(CNN),affording complex feature extraction.However,the local receptive fields in the CNN limit modeling long-range contextual relationships in semantic changes.Therefore,this work explores the great potential of Siamese Transformers in CD tasks and proposes a general CD model entitled STCD that relies on Swin Transformers.In the encoding process,pure Transformers without CNN are used to model the long-range context of semantic tokens,reducing computational overhead and improving model efficiency compared to current methods.During the decoding process,the 3D convolution block obtains the changing features in the time series and generates the predicted change map in the deconvolution layer with axial attention.Extensive experiments on three binary CD datasets and one semantic CD dataset demonstrate that the proposed STCD model outperforms several popular benchmark methods considering performance and the required parameters.Among the STCD variants,the F1-Score of the Base-STCD on the three binary CD datasets LEVIR,DSIFN,and SVCD reached 89.85%,54.72%,and 93.75%,respectively,and the mF1-Score and mIoU on the semantic CD dataset SECOND were 75.60%and 66.19%.
文摘Increased penetration of renewables for power generation has negatively impacted the dynamics of conventional fossil fuel-based power plants.The power plants operating on the base load are forced to cycle,to adjust to the fluctuating power demands.This results in an inefficient operation of the coal power plants,which leads up to higher operating losses.To overcome such operational challenge associated with cycling and to develop an optimal process control,this work analyzes a set of models for predicting power generation.Moreover,the power generation is intrinsically affected by the state of the power plant components,and therefore our model development also incorporates additional power plant process variables while forecasting the power generation.We present and compare multiple state-of-the-art forecasting data-driven methods for power generation to determine the most adequate and accurate model.We also develop an interpretable attention-based transformer model to explain the importance of process variables during training and forecasting.The trained deep neural network(DNN)LSTM model has good accuracy in predicting gross power generation under various prediction horizons with/without cycling events and outperforms the other models for long-term forecasting.The DNN memory-based models show significant superiority over other state-of-the-art machine learning models for short,medium and long range predictions.The transformer-based model with attention enhances the selection of historical data for multi-horizon forecasting,and also allows to interpret the significance of internal power plant components on the power generation.This newly gained insights can be used by operation engineers to anticipate and monitor the health of power plant equipment during high cycling periods.
