Qiangsheng Dai
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Qiangsheng Dai has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Jiansong Tang
Graduate School of Informatics, Osaka Metropolitan University, Osaka 559-8531, Japan
3 shared publications - Yingchi Mao
Hohai University
3 shared publications - Jinghui Zhang
Southeast University, Nanjing, China
2 shared publications - Libao Shi
Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen, China
2 shared publications Xuesong Huo
State Grid Jiangsu Electric Power Company
2 shared publications- Cui Wang
School of Electrical Engineering Jiangxi University of Water Resources and Electric Power Nanchang Jiangxi China
1 shared publication
Latest Publications
- Open Access
- 10 November 2026
- Published by MDPI AG in Electronics
- Vol. 15  (22) , 5201
An Ultra-Short-Term Probabilistic Load Forecasting Method Based on Joint Distribution Modeling
Ultra-short-term probabilistic load forecasting is of great significance for improving the safety and economy of the operation of new power systems. To address the problem that existing methods rarely focus on the dependence relationship between load forecast errors and the fluctuation characteristics of forecast output series, this paper proposes an ultra-short-term probabilistic load forecasting method based on joint distribution modeling. First, the Transformer model is used to perform high-precision point forecasting on historical load series, providing a reliable basis for subsequent uncertainty analysis. Second, based on an autoencoder, the forecast load series is reconstructed, and the reconstruction error is used as a deep fluctuation feature. The number of volatility patterns is determined through clustering-validity analysis, and the load scenarios are consequently divided into two representative fluctuation patterns, revealing the intrinsic relationship between the forecast-error distribution and load dynamic characteristics. Subsequently, for each pattern, a joint probability distribution model of load point forecast values and forecast errors is constructed separately; the Copula function is introduced to fit their dependence structure; the conditional error distribution is generated through inverse transformation; and finally, probabilistic prediction intervals are output according to confidence level requirements. Experiments on real-world load datasets show that the proposed method improves comprehensive accuracy metrics at different confidence levels.
- 10 November 2026
- Published by MDPI AG in Electronics
- Vol. 15  (22) , 5201
Open AccessAn Ultra-Short-Term Probabilistic Load Forecasting Method Based on Joint Distribution Modeling
Ultra-short-term probabilistic load forecasting is of great significance for improving the safety and economy of the operation of new power systems. To address the problem that existing methods rarely focus on the dependence relationship between load forecast errors and the fluctuation characteristics of forecast output series, this paper proposes an ultra-short-term probabilistic load forecasting method based on joint distribution modeling. First, the Transformer model is used to perform high-precision point forecasting on historical load series, providing a reliable basis for subsequent uncertainty analysis. Second, based on an autoencoder, the forecast load series is reconstructed, and the reconstruction error is used as a deep fluctuation feature. The number of volatility patterns is determined through clustering-validity analysis, and the load scenarios are consequently divided into two representative fluctuation patterns, revealing the intrinsic relationship between the forecast-error distribution and load dynamic characteristics. Subsequently, for each pattern, a joint probability distribution model of load point forecast values and forecast errors is constructed separately; the Copula function is introduced to fit their dependence structure; the conditional error distribution is generated through inverse transformation; and finally, probabilistic prediction intervals are output according to confidence level requirements. Experiments on real-world load datasets show that the proposed method improves comprehensive accuracy metrics at different confidence levels.
- 25 February 2025
- Published by Springer Nature
- p. 318-325
Offshore Wind Power Prediction Based on Improved K-MEANS and DCNN-LSTM-TRANSFORMER
- 25 February 2025
- Published by Springer Nature
- p. 318-325
Offshore Wind Power Prediction Based on Improved K-MEANS and DCNN-LSTM-TRANSFORMER
- Open Access
- 2 December 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14  (23) , 11217
STGPT2UGAN: Spatio-Temporal GPT-2 United Generative Adversarial Network for Wind Speed Prediction in Turbine Network
In a turbine network, an accurate wind speed prediction plays a key role in improving the efficiency of turbine operations. At present, Large Language Models (LLMs) have shown their strong capabilities in time sequence forecasting. Unfortunately, the existing LLM-based models face the following two difficulties in precise wind speed prediction: (1) These models fail to capture spatio-temporal correlations. (2) These models rely exclusively on supervised learning to train. Supervised learning updates parameters by optimizing a loss function, which focuses solely on improving accuracy on a specific regression or classification task rather than understanding the true data distribution. To tackle these difficulties, we devise a novel wind speed prediction architecture entitled Spatio-Temporal GPT-2 United Generative Adversarial Network (STGPT2UGAN). In particular, in order to model spatio-temporal correlations, we first design a Spatio-Temporal GPT-2 (STGPT-2), which comprises a Spatio-Temporal Block and the variant of GPT-2. Second, to improve the training of STGPT-2, a Generative Adversarial Network (GAN) is introduced to develop an adversarial training strategy. The strategy combines unsupervised training and supervised training in a complementary fashion. The unsupervised training promotes STGPT-2 to learn the distribution of true data through minimizing the adversarial loss. The supervised training intends to align the true values and the predicted values. We conduct the experiments on one real-world wind speed dataset. The experimental results verify that STGPT2UGAN outperforms the state-of-the-art benchmarks in terms of prediction precision.
- 2 December 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14  (23) , 11217
Open AccessSTGPT2UGAN: Spatio-Temporal GPT-2 United Generative Adversarial Network for Wind Speed Prediction in Turbine Network
In a turbine network, an accurate wind speed prediction plays a key role in improving the efficiency of turbine operations. At present, Large Language Models (LLMs) have shown their strong capabilities in time sequence forecasting. Unfortunately, the existing LLM-based models face the following two difficulties in precise wind speed prediction: (1) These models fail to capture spatio-temporal correlations. (2) These models rely exclusively on supervised learning to train. Supervised learning updates parameters by optimizing a loss function, which focuses solely on improving accuracy on a specific regression or classification task rather than understanding the true data distribution. To tackle these difficulties, we devise a novel wind speed prediction architecture entitled Spatio-Temporal GPT-2 United Generative Adversarial Network (STGPT2UGAN). In particular, in order to model spatio-temporal correlations, we first design a Spatio-Temporal GPT-2 (STGPT-2), which comprises a Spatio-Temporal Block and the variant of GPT-2. Second, to improve the training of STGPT-2, a Generative Adversarial Network (GAN) is introduced to develop an adversarial training strategy. The strategy combines unsupervised training and supervised training in a complementary fashion. The unsupervised training promotes STGPT-2 to learn the distribution of true data through minimizing the adversarial loss. The supervised training intends to align the true values and the predicted values. We conduct the experiments on one real-world wind speed dataset. The experimental results verify that STGPT2UGAN outperforms the state-of-the-art benchmarks in terms of prediction precision.
- Open Access
- 21 November 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13  (23) , 12561
Advanced Meteorological Hazard Defense Capability Assessment: Addressing Sample Imbalance with Deep Learning Approaches
With the rise in meteorological disasters, improving evaluation strategies for disaster response agencies is critical. This shift from expert scoring to data-driven approaches is challenged by sample imbalance in the data, affecting accurate capability assessments. This study proposes a solution integrating adaptive focal loss into the cross-entropy loss function to address sample distribution imbalances, facilitating nuanced evaluations. A key aspect of this solution is the Encoder-Adaptive-Focal deep learning model coupled with a custom training algorithm, adept at handling the data complexities of meteorological disaster response agencies. The model proficiently extracts and optimizes capability features from time series data, directing the evaluative focus toward more complex samples, thus mitigating sample imbalance issues. Comparative analysis with existing methods like UAE-NaiveBayes, UAE-SVM, and UAE-RandomForest illustrates the superior performance of our model in ability evaluation, positioning it as a robust tool for dynamic capability evaluation. This work aims to enhance disaster management strategies, contributing to mitigating the impacts of meteorological disasters.
- 21 November 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13  (23) , 12561
Open AccessAdvanced Meteorological Hazard Defense Capability Assessment: Addressing Sample Imbalance with Deep Learning Approaches
With the rise in meteorological disasters, improving evaluation strategies for disaster response agencies is critical. This shift from expert scoring to data-driven approaches is challenged by sample imbalance in the data, affecting accurate capability assessments. This study proposes a solution integrating adaptive focal loss into the cross-entropy loss function to address sample distribution imbalances, facilitating nuanced evaluations. A key aspect of this solution is the Encoder-Adaptive-Focal deep learning model coupled with a custom training algorithm, adept at handling the data complexities of meteorological disaster response agencies. The model proficiently extracts and optimizes capability features from time series data, directing the evaluative focus toward more complex samples, thus mitigating sample imbalance issues. Comparative analysis with existing methods like UAE-NaiveBayes, UAE-SVM, and UAE-RandomForest illustrates the superior performance of our model in ability evaluation, positioning it as a robust tool for dynamic capability evaluation. This work aims to enhance disaster management strategies, contributing to mitigating the impacts of meteorological disasters.
- Open Access
- 19 October 2023
- Published by Center of Biomass and Renewable Energy Scientia Academy in International Journal of Renewable Energy Development
- Vol. 12  (6) , 1104-1112
Photovoltaic power prediction based on sky images and tokens-to-token vision transformer
Photovoltaic (PV) power generation has high uncertainties due to the randomness and imbalance nature of solar energy and meteorological parameters. Hence, accurate PV power forecasts are essential in the operation of PV power plants (PVPP) for short-term dispatches and power generation schedules. In this paper, a new deep neural network structure based on vision transformer is proposed to combine sky images and Tokens-To-Token(T2T) for photovoltaic power prediction. The method uses an incremental tokenization module to aggregate neighboring image patches into tokens, which capture the local structural information of the clouds. Then, an efficient T2T-ViT backbone network is used to extract the global attentional relationships of the tokens for power prediction. In order to evaluate the performance of the proposed model, the method was compared with several deep learning architectures such as ResNet and GoogleNet on a dataset collected by the National Renewable Energy Laboratory in Colorado, USA. The results of power prediction were analysed using training loss, prediction error, and linear regression, and they show that the proposed method achieves higher prediction accuracy and lower error compared to the existing methods, especially in short- and ultra-short-term prediction. The paper demonstrates the potential of applying Transformer models to computer vision tasks for renewable energy forecasting. The results show that the proposed method achieves higher prediction accuracy and lower error than several deep learning architectures, such as ResNet and GoogleNet, especially in short- and ultra-short-term prediction.
- 19 October 2023
- Published by Center of Biomass and Renewable Energy Scientia Academy in International Journal of Renewable Energy Development
- Vol. 12  (6) , 1104-1112
Open AccessPhotovoltaic power prediction based on sky images and tokens-to-token vision transformer
Photovoltaic (PV) power generation has high uncertainties due to the randomness and imbalance nature of solar energy and meteorological parameters. Hence, accurate PV power forecasts are essential in the operation of PV power plants (PVPP) for short-term dispatches and power generation schedules. In this paper, a new deep neural network structure based on vision transformer is proposed to combine sky images and Tokens-To-Token(T2T) for photovoltaic power prediction. The method uses an incremental tokenization module to aggregate neighboring image patches into tokens, which capture the local structural information of the clouds. Then, an efficient T2T-ViT backbone network is used to extract the global attentional relationships of the tokens for power prediction. In order to evaluate the performance of the proposed model, the method was compared with several deep learning architectures such as ResNet and GoogleNet on a dataset collected by the National Renewable Energy Laboratory in Colorado, USA. The results of power prediction were analysed using training loss, prediction error, and linear regression, and they show that the proposed method achieves higher prediction accuracy and lower error compared to the existing methods, especially in short- and ultra-short-term prediction. The paper demonstrates the potential of applying Transformer models to computer vision tasks for renewable energy forecasting. The results show that the proposed method achieves higher prediction accuracy and lower error than several deep learning architectures, such as ResNet and GoogleNet, especially in short- and ultra-short-term prediction.
- Open Access
- 19 June 2023
- Published by Frontiers Media SA in Frontiers in Energy Research
- Vol. 11, 1204032
Spatio-temporal prediction for distributed PV generation system based on deep learning neural network model
To obtain higher accuracy of PV prediction to enhance PV power generation technology. This paper proposes a spatio-temporal prediction method based on a deep learning neural network model. Firstly, spatio-temporal correlation analysis is performed for 17 PV sites. Secondly, we compare CNN-LSTM with a single CNN or LSTM model trained on the same dataset. From the evaluation indexes such as loss map, regression map, RMSE, and MAE, the CNN-LSTM model that considers the strong correlation of spatio-temporal correlation among the 17 sites has better performance. The results show that our method has higher prediction accuracy.
- 19 June 2023
- Published by Frontiers Media SA in Frontiers in Energy Research
- Vol. 11, 1204032
Open AccessSpatio-temporal prediction for distributed PV generation system based on deep learning neural network model
To obtain higher accuracy of PV prediction to enhance PV power generation technology. This paper proposes a spatio-temporal prediction method based on a deep learning neural network model. Firstly, spatio-temporal correlation analysis is performed for 17 PV sites. Secondly, we compare CNN-LSTM with a single CNN or LSTM model trained on the same dataset. From the evaluation indexes such as loss map, regression map, RMSE, and MAE, the CNN-LSTM model that considers the strong correlation of spatio-temporal correlation among the 17 sites has better performance. The results show that our method has higher prediction accuracy.
- Open Access
- 15 June 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 14  (4) , 1-22
Noise-aware Local Model Training Mechanism for Federated Learning
As a new paradigm in training intelligent models, federated learning is widely used to train a global model without requiring local data to be uploaded from end devices. However, there are often mislabeled samples (i.e., noisy samples) in the dataset, which will cause the model update to deviate from the correct direction during the training process, thus reducing the convergence accuracy of the global model. Existing works employ noisy label correction techniques to reduce the impact of noisy samples on model updates by correcting labels; however, such methods necessitate the use of prior knowledge and additional communication costs, which cannot be directly applied to federated learning due to data privacy concerns and limited communication resources. Therefore, this paper proposes a noise-aware local model training method that corrects the noisy labels directly at the end device under the constraints of federated learning. By constructing a label correction model, a joint optimization problem is formally defined for optimizing both the label correction model and the client-side local training model (e.g., classification model). As a solution to this optimization problem, we propose a robustness training algorithm using label correction, along with a cross-validation data sampling algorithm that updates both models simultaneously. It is verified through experiments that the mechanism can effectively improve the model convergence accuracy on noisy datasets in federated learning scenarios.
- 15 June 2023
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Intelligent Systems and Technology
- Vol. 14  (4) , 1-22
Open AccessNoise-aware Local Model Training Mechanism for Federated Learning
As a new paradigm in training intelligent models, federated learning is widely used to train a global model without requiring local data to be uploaded from end devices. However, there are often mislabeled samples (i.e., noisy samples) in the dataset, which will cause the model update to deviate from the correct direction during the training process, thus reducing the convergence accuracy of the global model. Existing works employ noisy label correction techniques to reduce the impact of noisy samples on model updates by correcting labels; however, such methods necessitate the use of prior knowledge and additional communication costs, which cannot be directly applied to federated learning due to data privacy concerns and limited communication resources. Therefore, this paper proposes a noise-aware local model training method that corrects the noisy labels directly at the end device under the constraints of federated learning. By constructing a label correction model, a joint optimization problem is formally defined for optimizing both the label correction model and the client-side local training model (e.g., classification model). As a solution to this optimization problem, we propose a robustness training algorithm using label correction, along with a cross-validation data sampling algorithm that updates both models simultaneously. It is verified through experiments that the mechanism can effectively improve the model convergence accuracy on noisy datasets in federated learning scenarios.
- 22 May 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Power Electronics
- Vol. 38  (8) , 9933-9943
SVPWM Strategy With Minimum Common-Mode Voltage for Multilevel Converter Combining the Concept of the Nearest Level
Multilevel converters are widely used in medium- and high-voltage high-power industrial applications. Space vector pulsewidth modulation (SVPWM) is one of the very promising modulation strategies for multilevel converters. However, the SVPWM strategy is difficult to implement as the number of levels increases. The output common-mode voltage (CMV) of the converter is another problem in the implementation of SVPWM strategy. In this article, a new SVPWM strategy with minimum CMV for the multilevel converter is proposed by combining the concept of the nearest level. The proposed SVPWM strategy only utilizes simple arithmetic operations and logical judgments and completely avoids trigonometric functions and irrational number operations. Compared with the existing SVPWM methods, the proposed SVPWM strategy can quickly calculate the switching state with the minimum CMV in real time by the concept of the nearest level and avoids solving the nonhomogeneous linear equations in the process of calculating the switching state corresponding to the space vector. It can be extended to any level converter without additional computational burden. Finally, the proposed SVPWM strategy is verified with extensive simulation and experimental studies.
- 22 May 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Power Electronics
- Vol. 38  (8) , 9933-9943
SVPWM Strategy With Minimum Common-Mode Voltage for Multilevel Converter Combining the Concept of the Nearest Level
Multilevel converters are widely used in medium- and high-voltage high-power industrial applications. Space vector pulsewidth modulation (SVPWM) is one of the very promising modulation strategies for multilevel converters. However, the SVPWM strategy is difficult to implement as the number of levels increases. The output common-mode voltage (CMV) of the converter is another problem in the implementation of SVPWM strategy. In this article, a new SVPWM strategy with minimum CMV for the multilevel converter is proposed by combining the concept of the nearest level. The proposed SVPWM strategy only utilizes simple arithmetic operations and logical judgments and completely avoids trigonometric functions and irrational number operations. Compared with the existing SVPWM methods, the proposed SVPWM strategy can quickly calculate the switching state with the minimum CMV in real time by the concept of the nearest level and avoids solving the nonhomogeneous linear equations in the process of calculating the switching state corresponding to the space vector. It can be extended to any level converter without additional computational burden. Finally, the proposed SVPWM strategy is verified with extensive simulation and experimental studies.
- Open Access
- 20 April 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13  (8) , 5153
Research on Feature Extraction of Meteorological Disaster Emergency Response Capability Based on an RNN Autoencoder
Climate change has increased the frequency of various types of meteorological disasters in recent years. Finding the primary factors that limit the emergency response capability of meteorological disasters through the evaluation of that capability and proposing corresponding improvement measures in order to increase that capability is of great practical importance. The evaluation of meteorological disaster emergency response capability still has some issues. The majority of research methods use qualitative analysis, which makes it challenging to deal with fuzzy factors, leading to conclusions that are subjective and insufficiently rigorous. The evaluation models themselves are also complex and challenging to simulate and analyze, making it challenging to promote and use them in practice. Deep learning techniques have made it easier to collect and process large amounts of data, which has opened new avenues for advancement in the emergency management of weather-related disasters. In this paper, we suggest a Recurrent Neural Network (RNN)-based dynamic capability feature extraction method. The process of evaluation content determination and index selection is used to build a meteorological disaster emergency response capability evaluation index system before an encoder, based on the encoder–decoder architecture, is built for dynamic feature extraction. The RNN autoencoder deep learning ability dynamic rating method used in this paper has been shown through a series of experiments to be able to not only efficiently extract ability features from time series data and reduce the dimensionality of ability features, but also to reduce the focus of the ability evaluation model on simple and abnormal samples, concentrate the model learning on difficult samples, and have a higher accuracy. As a result, it is more suitable for the problem situation at evaluation of the disaster capability.
- 20 April 2023
- Published by MDPI AG in Applied Sciences
- Vol. 13  (8) , 5153
Open AccessResearch on Feature Extraction of Meteorological Disaster Emergency Response Capability Based on an RNN Autoencoder
Climate change has increased the frequency of various types of meteorological disasters in recent years. Finding the primary factors that limit the emergency response capability of meteorological disasters through the evaluation of that capability and proposing corresponding improvement measures in order to increase that capability is of great practical importance. The evaluation of meteorological disaster emergency response capability still has some issues. The majority of research methods use qualitative analysis, which makes it challenging to deal with fuzzy factors, leading to conclusions that are subjective and insufficiently rigorous. The evaluation models themselves are also complex and challenging to simulate and analyze, making it challenging to promote and use them in practice. Deep learning techniques have made it easier to collect and process large amounts of data, which has opened new avenues for advancement in the emergency management of weather-related disasters. In this paper, we suggest a Recurrent Neural Network (RNN)-based dynamic capability feature extraction method. The process of evaluation content determination and index selection is used to build a meteorological disaster emergency response capability evaluation index system before an encoder, based on the encoder–decoder architecture, is built for dynamic feature extraction. The RNN autoencoder deep learning ability dynamic rating method used in this paper has been shown through a series of experiments to be able to not only efficiently extract ability features from time series data and reduce the dimensionality of ability features, but also to reduce the focus of the ability evaluation model on simple and abnormal samples, concentrate the model learning on difficult samples, and have a higher accuracy. As a result, it is more suitable for the problem situation at evaluation of the disaster capability.