Kaihong Zheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Kaihong Zheng has a total of 58 co-authors. Here are the top co-authors by number of shared publications:
Sheng Li
China Southern Power Grid Digital Grid Group Company Ltd., Guangzhou, China
5 shared publicationsLukun Zeng
Digital Grid Research Institute, China Southern Power Grid, GuangZhou, China
4 shared publicationsShangli Zhou
Digital Grid Research Institute, China Southern Power Grid, GuangZhou, China
4 shared publicationsJingfeng Yang
China Southern Power Grid Company Ltd., Guangzhou, China
3 shared publicationsQihang Gong
Digital Grid Research Institute, China Southern Power Grid, GuangZhou, China
3 shared publicationsLei Zhong
Hainan Power Grid Company Ltd., China Southern Power Grid, Haikou, China
2 shared publications
Latest Publications
- Open Access
Anomaly Detection and Data Repair for Smart Meter Data in Smart Cities: A Comprehensive Review and Future Perspectives
Smart meters are the core terminals for distribution network data acquisition in smart cities, yet their collected data commonly suffer from quality issues caused by harsh operating environments, communication failures, hardware degradation, and human factors. This paper presents a systematic review of anomaly detection and data repair methods for smart meter data based on a critical analysis of many publications. First, we characterize five typical anomalies—sudden jumps, reading stagnation, reverse readings, pulse spikes, and gradual drifts—from physical root causes to data manifestations and provide unified mathematical definitions with explicit traceability to the existing literature. Additional anomaly types including meter replacement jumps, data duplication from retransmission, complete missing segments, and timestamp errors are also discussed to present a more complete picture of operational data quality challenges. Second, existing anomaly detection methods are systematically reviewed and classified into four categories—statistical, machine learning, deep learning, and dedicated time-series methods—with representative studies, quantitative performance metrics, and scenario-specific applicability examined for each. Third, data repair approaches are reviewed across four categories—traditional interpolation, matrix completion, generative models, and time-series prediction—with systematic comparison of their accuracy and limitations across different anomaly types and durations. Based on the synthesized evidence, we identify three cross-cutting structural limitations that persist across method categories: the performance ceiling of data-only detection without physical constraint embedding, the open-loop architecture that separates detection from repair and allows error propagation, and the exclusive reliance on statistical error metrics that fails to distinguish physically plausible repairs from those violating conservation laws. To address these gaps, we discuss a physics-guided integrated framework incorporating physical constraint embedding, joint anomaly diagnosis, scenario-adaptive repair, and posterior verification as a promising forward-looking direction. Finally, open challenges and future research directions are outlined, including parameter adaptation in unlabeled scenarios, multi-source data fusion for physical disambiguation, new power system extensions, explainable AI integration, edge-computing deployment, and standardized benchmark development. This review provides a comprehensive theoretical reference and technical roadmap for smart meter data quality research in the context of smart city energy systems.
Open AccessAnomaly Detection and Data Repair for Smart Meter Data in Smart Cities: A Comprehensive Review and Future Perspectives
Smart meters are the core terminals for distribution network data acquisition in smart cities, yet their collected data commonly suffer from quality issues caused by harsh operating environments, communication failures, hardware degradation, and human factors. This paper presents a systematic review of anomaly detection and data repair methods for smart meter data based on a critical analysis of many publications. First, we characterize five typical anomalies—sudden jumps, reading stagnation, reverse readings, pulse spikes, and gradual drifts—from physical root causes to data manifestations and provide unified mathematical definitions with explicit traceability to the existing literature. Additional anomaly types including meter replacement jumps, data duplication from retransmission, complete missing segments, and timestamp errors are also discussed to present a more complete picture of operational data quality challenges. Second, existing anomaly detection methods are systematically reviewed and classified into four categories—statistical, machine learning, deep learning, and dedicated time-series methods—with representative studies, quantitative performance metrics, and scenario-specific applicability examined for each. Third, data repair approaches are reviewed across four categories—traditional interpolation, matrix completion, generative models, and time-series prediction—with systematic comparison of their accuracy and limitations across different anomaly types and durations. Based on the synthesized evidence, we identify three cross-cutting structural limitations that persist across method categories: the performance ceiling of data-only detection without physical constraint embedding, the open-loop architecture that separates detection from repair and allows error propagation, and the exclusive reliance on statistical error metrics that fails to distinguish physically plausible repairs from those violating conservation laws. To address these gaps, we discuss a physics-guided integrated framework incorporating physical constraint embedding, joint anomaly diagnosis, scenario-adaptive repair, and posterior verification as a promising forward-looking direction. Finally, open challenges and future research directions are outlined, including parameter adaptation in unlabeled scenarios, multi-source data fusion for physical disambiguation, new power system extensions, explainable AI integration, edge-computing deployment, and standardized benchmark development. This review provides a comprehensive theoretical reference and technical roadmap for smart meter data quality research in the context of smart city energy systems.
- 12 July 2026
- Published by Springer Nature
- p. 559-572
Label Correlation Embedding Based Graph Attention Neural Network for Multi-instance Multi-label Learning
- 12 July 2026
- Published by Springer Nature
- p. 559-572
Label Correlation Embedding Based Graph Attention Neural Network for Multi-instance Multi-label Learning
- Open Access
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 18332-18342
Federated Contrastive Graph Attention Neural Network With Correlation Embedding for Noisy Multi-Label Classification
Multi-label image classification in privacy-sensitive domains faces several challenges, particularly in multi-label electricity scene classification tasks where equipment wear and similar factors introduce label noise. The fundamental limitation lies in most FL algorithms’ implicit assumption of clean local datasets, making them inherently susceptible to performance degradation when faced with the realistic challenges of noisy labels in distributed learning scenarios. Moreover, in current federated learning methods for noisy multi-label classification, few approaches consider modeling the correlations between labels to assist in the classification task of noisy labels. Specifically, they rarely address the differences in correlations between noisy labels and true labels, as well as between true labels themselves. Additionally, when noisy labels are present on the client side, the model is prone to overfitting to these noisy labels, which negatively impacts its recognition capability. Under this situation, this paper proposes a federated contrastive graph attention neural network with correlation embedding (FedCGAT) for noisy multi-label classification. FedCGAT alleviates the impact of noisy labels by introducing consistency constraints on client models to build consistent feature representations. Specifically, FedCGAT designs client-side graph structures to capture label semantic information, leveraging label correlations to provide corrective guidance for noisy annotations. Furthermore, it designs a dual-branch contrastive learning framework integrating local and global models to construct model support for label denoising. This dual-pronged approach jointly learns discriminative features from both label semantics and model representations, thereby achieving robust noisy multi-label classification. Experimental validation on federated electricity scene image datasets confirms the effectiveness of the proposed model.
- 15 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 18332-18342
Open AccessFederated Contrastive Graph Attention Neural Network With Correlation Embedding for Noisy Multi-Label Classification
Multi-label image classification in privacy-sensitive domains faces several challenges, particularly in multi-label electricity scene classification tasks where equipment wear and similar factors introduce label noise. The fundamental limitation lies in most FL algorithms’ implicit assumption of clean local datasets, making them inherently susceptible to performance degradation when faced with the realistic challenges of noisy labels in distributed learning scenarios. Moreover, in current federated learning methods for noisy multi-label classification, few approaches consider modeling the correlations between labels to assist in the classification task of noisy labels. Specifically, they rarely address the differences in correlations between noisy labels and true labels, as well as between true labels themselves. Additionally, when noisy labels are present on the client side, the model is prone to overfitting to these noisy labels, which negatively impacts its recognition capability. Under this situation, this paper proposes a federated contrastive graph attention neural network with correlation embedding (FedCGAT) for noisy multi-label classification. FedCGAT alleviates the impact of noisy labels by introducing consistency constraints on client models to build consistent feature representations. Specifically, FedCGAT designs client-side graph structures to capture label semantic information, leveraging label correlations to provide corrective guidance for noisy annotations. Furthermore, it designs a dual-branch contrastive learning framework integrating local and global models to construct model support for label denoising. This dual-pronged approach jointly learns discriminative features from both label semantics and model representations, thereby achieving robust noisy multi-label classification. Experimental validation on federated electricity scene image datasets confirms the effectiveness of the proposed model.
- Open Access
Enhanced Short-Term Load Forecasting Based on Adaptive Residual Fusion of Autoformer and Transformer
Accurate short-term electricity load forecasting (STELF) is essential for grid scheduling and low-carbon smart grids. However, load exhibits multi-timescale periodicity and non-stationary fluctuations, making STELF highly challenging for existing models. To address this challenge, an Autoformer–Transformer residual fusion network (ATRFN) is proposed in this paper. A dynamic weighting mechanism is applied to combine the outputs of Autoformer and Transformer through residual connections. In this way, lightweight result-level fusion is enabled without modifications to either architecture. In experimental validations on real-world load datasets, the proposed ATRFN model achieves notable performance gains over single STELF models. For univariate STELF, the ATRFN model reduces forecasting errors by 11.94% in mean squared error (MSE), 10.51% in mean absolute error (MAE), and 7.99% in mean absolute percentage error (MAPE) compared with the best single model. In multivariate experiments, it further decreases errors by at least 5.22% in MSE, 2.77% in MAE, and 2.85% in MAPE, demonstrating consistent improvements in predictive accuracy.
Open AccessEnhanced Short-Term Load Forecasting Based on Adaptive Residual Fusion of Autoformer and Transformer
Accurate short-term electricity load forecasting (STELF) is essential for grid scheduling and low-carbon smart grids. However, load exhibits multi-timescale periodicity and non-stationary fluctuations, making STELF highly challenging for existing models. To address this challenge, an Autoformer–Transformer residual fusion network (ATRFN) is proposed in this paper. A dynamic weighting mechanism is applied to combine the outputs of Autoformer and Transformer through residual connections. In this way, lightweight result-level fusion is enabled without modifications to either architecture. In experimental validations on real-world load datasets, the proposed ATRFN model achieves notable performance gains over single STELF models. For univariate STELF, the ATRFN model reduces forecasting errors by 11.94% in mean squared error (MSE), 10.51% in mean absolute error (MAE), and 7.99% in mean absolute percentage error (MAPE) compared with the best single model. In multivariate experiments, it further decreases errors by at least 5.22% in MSE, 2.77% in MAE, and 2.85% in MAPE, demonstrating consistent improvements in predictive accuracy.
- Open Access
- 4 November 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-14
Federated asynchronous graph attention network with structural semantic embedding for multi-label graph classification
Federated Learning (FL) provides a privacy-preserving framework for training graph neural networks (GNNs) in privacy-sensitive scenarios. However, traditional FL-GNN approaches often focus on addressing data distribution inconsistencies across clients from a purely data-centric viewpoint, overlooking the critical role of label semantics. Incorporating label semantic information, however, significantly enhances a model’s performance in multi-label classification tasks. Moreover, the performance of FL-GNNs is constrained by two key challenges related to heterogeneity: intra-client heterogeneity in graph representations and inter-client heterogeneity across distributed graphs. Unfortunately, few FL methods effectively handle discrepancies in both label distributions and graph heterogeneity across different clients. To address this gap, we introduce the federated asynchronous graph attention network with structural semantic embedding for multi-label classification (FasSGAT). FasSGAT primarily contains: (1) client-specific label semantic embedding modules that learn feature encodings from constructed label-semantic distribution graphs, (2) the integration of these encodings into the backbone multi-label classifier, along with specially designed structure-sensitive spectral features to mitigate client-side heterogeneity, and (3) a novel structure-sensitive asynchronous aggregation mechanism at the server level that uses the graph spectral features to construct a global model and address graph heterogeneity. Our experimental results on multi-label benchmarks show that FasSGAT outperforms traditional FL methods across various evaluation metrics.
- 4 November 2025
- Published by Springer Nature in Scientific Reports
- Vol. 15 (1) , 1-14
Open AccessFederated asynchronous graph attention network with structural semantic embedding for multi-label graph classification
Federated Learning (FL) provides a privacy-preserving framework for training graph neural networks (GNNs) in privacy-sensitive scenarios. However, traditional FL-GNN approaches often focus on addressing data distribution inconsistencies across clients from a purely data-centric viewpoint, overlooking the critical role of label semantics. Incorporating label semantic information, however, significantly enhances a model’s performance in multi-label classification tasks. Moreover, the performance of FL-GNNs is constrained by two key challenges related to heterogeneity: intra-client heterogeneity in graph representations and inter-client heterogeneity across distributed graphs. Unfortunately, few FL methods effectively handle discrepancies in both label distributions and graph heterogeneity across different clients. To address this gap, we introduce the federated asynchronous graph attention network with structural semantic embedding for multi-label classification (FasSGAT). FasSGAT primarily contains: (1) client-specific label semantic embedding modules that learn feature encodings from constructed label-semantic distribution graphs, (2) the integration of these encodings into the backbone multi-label classifier, along with specially designed structure-sensitive spectral features to mitigate client-side heterogeneity, and (3) a novel structure-sensitive asynchronous aggregation mechanism at the server level that uses the graph spectral features to construct a global model and address graph heterogeneity. Our experimental results on multi-label benchmarks show that FasSGAT outperforms traditional FL methods across various evaluation metrics.
- Open Access
- 14 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 5809-5820
Weighted Correlation Attention Network-Based Dual-Phase Federated Learning Framework for Image Recognition
The recognition task of privacy-sensitive images requires effective federated learning (FL) models. However, there are challenges such as inconsistent data distributions, as well as the impact of different class correlations and class imbalance for classification across regions. Existing research on explicitly addressing class correlation and class imbalance within FL frameworks remains limited. Additionally, due to the heterogeneity and class imbalance across different clients, the global model’s parameter aggregation process faces issues of inconsistency, resulting in a significant discrepancy between the global model parameters and parts of local models’ parameters, which leads to poor classification performance on these clients’ local models. To address these challenges, we propose a novel dual-phase federated learning framework, Fd-WCAT, which incorporates a weighted correlation attention network for multi-label electricity image recognition. In this framework, each client trains a local model, constructing masked label correlation graphs. Then, the extracted label correlation embedding features are extracted from the graphs for a class imbalance-weighted classifier. To tackle the parameter inconsistency between local and global models during training, Fd-WCAT designs a dual-phase weighted loss function based on global-local parameter regularization. Each client computes its class imbalance coefficient and sends the local model parameters to the server. On the server side, local models are clustered into T groups to ensure similarity in model parameters within each group. Intra-group aggregation is performed to generate T temporary models with minimal inconsistencies. Then, imbalance weights for the prototype models are calculated based on the average imbalance coefficient of each group, and a global model is generated through imbalance weighted aggregation. Experimental results on multi-label benchmarks show that Fd-WCAT outperforms existing baseline models.
- 14 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 5809-5820
Open AccessWeighted Correlation Attention Network-Based Dual-Phase Federated Learning Framework for Image Recognition
The recognition task of privacy-sensitive images requires effective federated learning (FL) models. However, there are challenges such as inconsistent data distributions, as well as the impact of different class correlations and class imbalance for classification across regions. Existing research on explicitly addressing class correlation and class imbalance within FL frameworks remains limited. Additionally, due to the heterogeneity and class imbalance across different clients, the global model’s parameter aggregation process faces issues of inconsistency, resulting in a significant discrepancy between the global model parameters and parts of local models’ parameters, which leads to poor classification performance on these clients’ local models. To address these challenges, we propose a novel dual-phase federated learning framework, Fd-WCAT, which incorporates a weighted correlation attention network for multi-label electricity image recognition. In this framework, each client trains a local model, constructing masked label correlation graphs. Then, the extracted label correlation embedding features are extracted from the graphs for a class imbalance-weighted classifier. To tackle the parameter inconsistency between local and global models during training, Fd-WCAT designs a dual-phase weighted loss function based on global-local parameter regularization. Each client computes its class imbalance coefficient and sends the local model parameters to the server. On the server side, local models are clustered into T groups to ensure similarity in model parameters within each group. Intra-group aggregation is performed to generate T temporary models with minimal inconsistencies. Then, imbalance weights for the prototype models are calculated based on the average imbalance coefficient of each group, and a global model is generated through imbalance weighted aggregation. Experimental results on multi-label benchmarks show that Fd-WCAT outperforms existing baseline models.
- Open Access
- 5 August 2025
- Published by MDPI AG in Journal of Low Power Electronics and Applications
- Vol. 15 (3) , 46
Federated Multi-Stage Attention Neural Network for Multi-Label Electricity Scene Classification
Privacy-sensitive electricity scene classification requires robust models under data localization constraints, making federated learning (FL) a suitable framework. Existing FL frameworks face two critical challenges in multi-label electricity scene classification: (1) Label correlations and their strengths significantly impact classification performance. (2) Electricity scene data and labels show distributional inconsistencies across regions. However, current FL frameworks lack explicit modeling of label correlation strengths, and locally trained regional models naturally capture these differences, leading to regional differences in their model parameters. In this scenario, the server’s standard single-stage aggregation often over-averages the global model’s parameters, reducing its discriminative ability. To address these issues, we propose FMMAN, a federated multi-stage attention neural network for multi-label electricity scene classification. The main contributions of this FMMAN lie in label correlation learning and the stepwise model aggregation. It splits the client–server interaction into multiple stages: (1) Clients train models locally to encode features and label correlation strengths after receiving the server’s initial model. (2) The server clusters these locally trained models into K groups to ensure that models within a group have more consistent parameters and generates K prototype models via intra-group aggregation to reduce over-averaging. The K models are then distributed back to the clients. (3) Clients refine their models using the K prototypes with contrastive group-specific consistency regularization to further mitigate over-averaging, and sends the refined model back to the server. (4) Finally, the server aggregates the models into a global model. Experiments on multi-label benchmarks verify that FMMAN outperforms baseline methods.
- 5 August 2025
- Published by MDPI AG in Journal of Low Power Electronics and Applications
- Vol. 15 (3) , 46
Open AccessFederated Multi-Stage Attention Neural Network for Multi-Label Electricity Scene Classification
Privacy-sensitive electricity scene classification requires robust models under data localization constraints, making federated learning (FL) a suitable framework. Existing FL frameworks face two critical challenges in multi-label electricity scene classification: (1) Label correlations and their strengths significantly impact classification performance. (2) Electricity scene data and labels show distributional inconsistencies across regions. However, current FL frameworks lack explicit modeling of label correlation strengths, and locally trained regional models naturally capture these differences, leading to regional differences in their model parameters. In this scenario, the server’s standard single-stage aggregation often over-averages the global model’s parameters, reducing its discriminative ability. To address these issues, we propose FMMAN, a federated multi-stage attention neural network for multi-label electricity scene classification. The main contributions of this FMMAN lie in label correlation learning and the stepwise model aggregation. It splits the client–server interaction into multiple stages: (1) Clients train models locally to encode features and label correlation strengths after receiving the server’s initial model. (2) The server clusters these locally trained models into K groups to ensure that models within a group have more consistent parameters and generates K prototype models via intra-group aggregation to reduce over-averaging. The K models are then distributed back to the clients. (3) Clients refine their models using the K prototypes with contrastive group-specific consistency regularization to further mitigate over-averaging, and sends the refined model back to the server. (4) Finally, the server aggregates the models into a global model. Experiments on multi-label benchmarks verify that FMMAN outperforms baseline methods.
- Open Access
Multi-Scale Graph Attention Network Based on Encoding Decomposition for Electricity Consumption Prediction
Accurate electricity consumption forecasting is essential for power scheduling. In short-term forecasting, electricity consumption data exhibit periodic patterns, as well as fluctuations associated with production events. Traditional forecasting methods typically focus on sequential features of the data, which may lead to an over-smoothing issue for the fluctuations. In practice, the fluctuations of electricity consumption associated with these events tend to follow recognizable patterns. By emphasizing the impact of these experiential electricity consumption fluctuations on the current prediction process, we can capture the volatility variations to alleviate the over-smoothing problem. To this end, we propose an encoding decomposition-based multi-scale graph neural network (CMNN). The CMNN starts by decomposing the electricity data into various components. For the high-order components that exhibit approximate periodic behavior, the CMNN designs a Multi-scale Bi-directional Long Short-Term Memory (MBLSTM) network for fitting and prediction. For the low-order components that exhibit fluctuations, the CMNN transforms these components from one-dimensional time series into a two-dimensional low-order component graph to model the volatility of the low-order components, and proposes a Gaussian Graph Auto-Encoder to forecast the low-order components. Finally, the CMNN combines the predicted components to produce the final electricity consumption prediction. Experiments demonstrate that the CMNN enhances the accuracy of electricity consumption predictions.
Open AccessMulti-Scale Graph Attention Network Based on Encoding Decomposition for Electricity Consumption Prediction
Accurate electricity consumption forecasting is essential for power scheduling. In short-term forecasting, electricity consumption data exhibit periodic patterns, as well as fluctuations associated with production events. Traditional forecasting methods typically focus on sequential features of the data, which may lead to an over-smoothing issue for the fluctuations. In practice, the fluctuations of electricity consumption associated with these events tend to follow recognizable patterns. By emphasizing the impact of these experiential electricity consumption fluctuations on the current prediction process, we can capture the volatility variations to alleviate the over-smoothing problem. To this end, we propose an encoding decomposition-based multi-scale graph neural network (CMNN). The CMNN starts by decomposing the electricity data into various components. For the high-order components that exhibit approximate periodic behavior, the CMNN designs a Multi-scale Bi-directional Long Short-Term Memory (MBLSTM) network for fitting and prediction. For the low-order components that exhibit fluctuations, the CMNN transforms these components from one-dimensional time series into a two-dimensional low-order component graph to model the volatility of the low-order components, and proposes a Gaussian Graph Auto-Encoder to forecast the low-order components. Finally, the CMNN combines the predicted components to produce the final electricity consumption prediction. Experiments demonstrate that the CMNN enhances the accuracy of electricity consumption predictions.
- Open Access
- 29 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 148665-148675
Multivariate Extreme Learning Machine Based AutoEncoder for Electricity Consumption Series Clustering
Multivariate electricity consumption series clustering can reflect the trend of power consumption changes in the past time period, which can provide reliable guidance for electricity production. The dimensionality reduction-based method is an effective technology to address this problem, which obtains the low-dimensional features of each variate or all variates for multivariate time series clustering. However, most existing dimensionality reduction-based methods ignore the joint learning of the common representations and the variable-based representations. In this paper, we build a multivariate extreme learning machine based autoencoder model for electricity consumption clustering (MELM-EC), which performs common representation learning and variable-based representation learning simultaneously. MELM-EC maps the common representation and multiple variable-based representations to the original multivariate time series and computes the common output weights within a few iterations. Experimental results on realistic multivariate time series datasets and multivariate electricity consumption series datasets demonstrate the effectiveness of the proposed MELM-EC model.
- 29 October 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 9, 148665-148675
Open AccessMultivariate Extreme Learning Machine Based AutoEncoder for Electricity Consumption Series Clustering
Multivariate electricity consumption series clustering can reflect the trend of power consumption changes in the past time period, which can provide reliable guidance for electricity production. The dimensionality reduction-based method is an effective technology to address this problem, which obtains the low-dimensional features of each variate or all variates for multivariate time series clustering. However, most existing dimensionality reduction-based methods ignore the joint learning of the common representations and the variable-based representations. In this paper, we build a multivariate extreme learning machine based autoencoder model for electricity consumption clustering (MELM-EC), which performs common representation learning and variable-based representation learning simultaneously. MELM-EC maps the common representation and multiple variable-based representations to the original multivariate time series and computes the common output weights within a few iterations. Experimental results on realistic multivariate time series datasets and multivariate electricity consumption series datasets demonstrate the effectiveness of the proposed MELM-EC model.
- 25 September 2021
- Published by Springer Nature in Frontiers of Computer Science
- Vol. 16 (2) , 1-11
EcoVis: visual analysis of industrial-level spatio-temporal correlations in electricity consumption
- 25 September 2021
- Published by Springer Nature in Frontiers of Computer Science
- Vol. 16 (2) , 1-11
EcoVis: visual analysis of industrial-level spatio-temporal correlations in electricity consumption