This article proposes a novel spatial–temporal resilience assessment scheme for the distribution system (DS) to address the issue of overoptimistic and potentially misleading assessment due to the imprecision of a single typhoon wind speed model and the neglect of the coupled impact of typhoon rainstorm. First, a dynamic weighted iterative algorithm (DWIA) is proposed for accurate modeling of the typhoon wind speed, and a DWIA-based equivalent wind speed (EWS) model is further proposed for improving the accuracy of wind speed calculation under the coupling effect of typhoon rainstorm. Then, EWS-based uncertain failure probability models are developed for passive and active components, such as distribution lines, poles, and photovoltaics, and the adverse impacts of typhoon coupled with rainstorm events (TCREs) on the outputs of active components are considered. Furthermore, a set of normalized resilience metrics is defined, and sequential Monte Carlo simulation is adopted to evaluate the DS resilience against the TCRE as it travels inland. Finally, the modified IEEE 33-bus system and the actual distribution system from Southern China affected by the TCRE are both tested and analyzed to validate the effectiveness of the proposed scheme. The numerical results show that the proposed scheme can effectively quantify the adverse impact of the TCRE on the DS, and provide assistance for the proactive resilience control of the DS.

In modern Internet of electric energy, i.e., networked power systems, data-driven schemes based on advanced machine learning methods have shown high potential in system emergency stability control, e.g., undervoltage load shedding (UVLS) against the short-term voltage stability (SVS) problem. However, how to efficiently and adaptively select the most effective UVLS sites for online SVS enhancement is still a challenging task. Faced with this issue, this paper develops an intelligent short-term voltage trajectory sensitivity index (VTSI) prediction scheme for adaptive UVLS site selection. Specifically, the scheme is realized by designing a powerful structure-aware recurrent learning machine (SRLM), which systematically combines the emerging graph convolutional network (GCN) with the recurrent long short-term memory algorithm. By doing so, the SRLM is not only fully aware of the non-Euclidean structure of the power grid, but also capable of amply capturing temporal features during SVS dynamics. Consequently, it manages to implement efficient and precise VTSI prediction, thereby reliably identifying critical UVLS sites in various scenarios. Numerical case studies on the Nordic test system illustrate the efficacy of the proposed scheme.

In industrial informatics-enabled smart grids, machine learning approaches have exhibited high potential in data-driven electricity theft detection (ETD), whereas none of the existing studies pay sufficient attention to the high costs of manually labeling massive sensing data during learning data preparation. To address this defect, this article develops a cost-effective data-driven ETD approach that significantly reduces the data labeling costs without sacrificing the reliability of ETD. Specifically, the approach is systematically realized via an intelligent deep active learning (DAL) scheme. By seamlessly incorporating convolutional neural network (CNN) learning with Monte Carlo dropout-based Bayesian active query, the DAL scheme efficiently selects the most valuable instances for ETD model training. In this way, the proposed approach is able to derive a reliable CNN-based ETD model with limited labeled learning instances, thus largely reducing the data labeling costs. Experimental test results on an actual ETD dataset provided by the State Grid Corporation of China extensively illustrate the efficacy of the proposed approach.

With the help of neural network-based representation learning, significant progress has been recently made in data-driven online dynamic stability assessment (DSA) of complex electric power systems. However, without sufficient attention to diverse data loss conditions in practice, the existing data-driven DSA solutions' performance could be largely degraded due to practical defective input data. To address this problem, this work develops a robust representation learning approach to enhance DSA performance against multiple input data loss conditions in practice. Specifically, focusing on the short-term voltage stability (SVS) issue, an ensemble representation learning scheme (ERLS) is carefully designed to achieve data loss-tolerant online SVS assessment: 1) based on an efficient data masking technique, various missing data conditions are handled and augmented in a unified manner for lossy learning dataset preparation; 2) the emerging spatial-temporal graph convolutional network (STGCN) is leveraged to derive multiple diversified base learners with strong capability in SVS feature learning and representation; and 3) with massive SVS scenarios deeply grouped into a number of clusters, these STGCN-enabled base learners are distinctly assembled for each cluster via multilinear regression (MLR) to realize ensemble SVS assessment. Such a divide-and-conquer ensemble strategy results in highly robust SVS assessment performance when faced with various severe data loss conditions. Numerical tests on the benchmark Nordic test system illustrate the efficacy of the proposed approach.

Smart grid is a typical application of industrial cyber-physical systems (ICPS) in the electric power industry. Due to the exposure to different kinds of uncertainties and unpredictable faults, how to reliably assess the short-term voltage stability (SVS) of smart grids to prevent the occurrence of large-scale blackouts is still of primary concern. To tackle this challenging problem, this article develops a novel machine learning scheme to achieve accurate and interpretable online SVS assessment in two steps. First, it utilizes time-series shapelet transform to extract key dynamics and convert the postfault time series into flat features. Second, it designs a graph convolutional network (GCN) to incorporate these features with topology information for SVS assessment. The GCN explores the spatial-temporal dynamics of power system via graph convolution and introduces a system layer to derive the final assessment result. Compared with conventional methods, this novel scheme makes full use of the spatial-temporal information in SVS dynamics, resulting in higher assessment accuracy and stronger adaptability. Besides, it is capable of discovering certain valuable underlying rules and patterns related to SVS. Test results on the IEEE 39-bus system and real-world Guangdong Power Grid in South China verify the effectiveness of the proposed scheme. Note to Practitioners—To achieve accurate and interpretable online short-term voltage stability (SVS) assessment in the challenging environment of smart grids, this article develops a novel machine learning scheme with full consideration of the spatial-temporal information in SVS dynamics. First, it utilizes the time series shapelet transform to convert the postfault time series into flat features. Second, it designs a graph convolutional network (GCN) to incorporate these features with topology information. The full consideration of spatial-temporal information in SVS dynamics can improve the assessment accuracy, and the integration of topology in the scheme can promote its adaptability to topology changes. Apart from the decent performances under changeable environments, the proposed scheme can provide certain valuable underlying rules and patterns related to SVS. Therefore, not only the proposed scheme for SVS assessment can work well in the practical challenging environment of smart grids, but also it helps the dispatchers in smart grids better understand and trust the proposed SVS assessment scheme.

In Internet of Things (IoT)-enabled modern power grids, advanced IoT devices, e.g., synchronous phasor measurement units (PMUs), have been widely deployed to closely monitor the grids’ states and dynamics. In practice, however, PMU measurements are often contaminated with anomalous (low-quality) data, e.g., data spikes, unchanged data, data losses/dropouts, and high-level data errors. To ensure the reliability of various PMU data-based applications, it is imperative to efficiently implement PMU data anomaly identification (PDAI). Focusing on performing online PDAI in a cost-effective way, this article develops an intelligent data-driven PDAI approach for practical power grids. Given the defect that the majority of the existing data-driven PDAI efforts necessitate costly domain expertise-based data annotation to start offline learning, the PDAI approach in this article is realized by designing an auto-starting semisupervised learning (SSL) scheme that automatically starts to learn from totally unlabeled PMU data. First, on the basis of the inherent spatial–temporal correlations in regional PMU measurements, sequential PMU data acquired from a specific power grid are characterized in a discriminative manner by profiling their spatial–temporal nearest neighbors (STNNs). With the exploration of the discriminability of STNN profiles, part of the obviously anomalous/normal data is reliably labeled on the basis of a statistical prior knowledge-based rule. Such an STNN-based preprocessing technique for partial data labeling enables the desirable auto-starting functionality of the whole SSL scheme. Then, taking both labeled and unlabeled data as inputs, a recurrent SSL machine for PDAI is efficiently built in two steps, i.e., unsupervised pretraining and supervised fine-tuning. Numerical test results with simulated PMU data from the Nordic test system and actual PMU data from two practical power grids illustrate the excellent PDAI performances of the proposed approach during the online application.

In modern Internet of Things-enabled smart grids, while existing data-driven efforts have made remarkable progress in wind turbine (WT) condition monitoring, the majority of them overlook potential data label errors w.r.t. WT fault information. To address this inadequacy, this article develops a two-stage automatic data label calibration (ADLC) approach via cost-effective time series (TS) data analytics. First, by profiling TS subsequence nearest neighbors (SNNs), most of the data labels are credibly calibrated by exploring the inherent temporal similarities within measured WT dynamics. To enhance the reliability of label calibration, ensemble labeling decisions are robustly made based on the diversity of multiplex variables. Then, for the remaining data labels not ascertained yet, a$k$SNN profiling method inspired by classical$k$nearest neighbor search is introduced to efficiently determine them from a complementary perspective. With no need for expensive domain expertise or computationally costly training procedures, the proposed approach can reliably calibrate WT fault information with a high efficiency, which makes it suitable for deploying into existing WT monitoring systems. Experimental test results with field data acquired from an actual WT illustrate the superior performances of the proposed approach.

For data-driven dynamic stability assessment (DSA) in power systems, learning cases collected from actual historical records appear to be more reliable than those obtained from numerical simulations with an inevitable reality gap. However, due to the scarceness of transient events in practical systems, historical case sets generally encounter the small sample size and class-imbalance problems. To tackle these challenging issues, this article proposes a novel data/model jointly driven framework to generate high-quality cases for power system DSA applications. Model-driven numerical simulations are first utilized for rough case generation, based upon which case refinement is then intelligently carried out via cycle generative adversarial network (CycleGAN) learning. In this data-driven manner, the CycleGAN is able to produce refined cases highly resembling actual historical ones. A long short-term memory-based semisupervised learning scheme is further designed to reliably label all the refined cases. Numerical tests are comprehensively carried out on the realistic Guangdong Power Grid in South China. With only a small and skewed historical case set initially provided, the proposed framework is able to generate highly realistic cases to augment the set and mitigate the class-imbalance issue. These synthetic cases further help derive a more discerning DSA model, which contributes to enhanced reliability and adaptability of online DSA in practical power grids.

In modern smart grids deployed with various advanced sensors, e.g., phasor measurement units (PMUs), bad (anomalous) measurements are always inevitable in practice. Considering the imperative need for filtering out potential bad data, this article develops a novel online bad PMU data detection (BPDD) approach for regional phasor data concentrators (PDCs) by sufficiently exploring spatial–temporal correlations. With no need for costly data labeling or iterative learning, it performs model-free, label-free, and noniterative BPDD in power grids from a new data-driven perspective of spatial–temporal nearest neighbor (STNN) discovery. Specifically, spatial–temporally correlated regional measurements acquired by PMUs are first gathered as a spatial–temporal time-series (TS) profile. Afterward, TS subsequences contaminated with bad PMU data are identified by characterizing anomalous STNNs. To make the whole approach competent in processing online streaming PMU data, an efficient strategy for accelerating STNN discovery is carefully designed. Different from existing data-driven BPDD solutions requiring either costly offline data set preparation/training or computationally intensive online optimization, it can be implemented in a highly cost-effective way, thereby being more applicable and scalable in practical contexts. Numerical test results on the Nordic test system and the realistic China Southern Power Grid demonstrate the reliability, efficiency, and scalability of the proposed approach in practical online monitoring.

While various promising phasor measurement unit (PMU) data-driven applications have been developed for modern smart grids, how to improve the overall PMU data quality to ensure the reliability of these applications in practice still remains an open issue. Considering the challenging task of missing PMU data correction (MPDC) in practical complicated and noisy measurement contexts, this article develops a novel spatiotemporal correlation learning scheme (SCLS) for online MPDC in smart grids. In particular, the SCLS is strategically realized with two successive modules. First, from four complementary spatiotemporal perspectives, statistical missing data imputation is carried out to derive initial correction results in noisy contexts. Second, a well-designed deep learning architecture with the integration of convolutional neural network (CNN) and residual learning techniques is introduced to refine the correction results. With the help of these two modules, the SCLS is capable of performing precise MPDC for regional PMU measurements as well as filtering out potential noises. Extensive numerical test results on the IEEE 39-bus test system and the real-world Guangdong power grid in South China demonstrate the efficacy of the SCLS in practical complicated contexts.

Focusing on fully learning intrinsic spatial and temporal dependencies from smart grids' complicated transients in a computationally efficient way, this article develops an intelligent machine learning approach for online short-term voltage stability (SVS) assessment. Based on static network information and dynamic system responses, spatial correlations are first comprehensively described from both model-based and data-based viewpoints. Such correlations are further formulated as spatial attention factors, which are leveraged to carefully rectify multiple transient trajectories. Taking the rectified trajectories as inputs, the long short-term memory based deep recurrent neural network (RNN) algorithm is employed to learn sequential SVS features. In this way, the RNN learning procedure is comprehensively guided by both spatial and temporal information, thereby deriving a highly reliable and robust classification model for online SVS assessment. Extensive numerical tests on the Nordic test system and the realistic Guangdong Power Grid in South China illustrate the superior reliability, scalability, and applicability of the proposed approach over existing methods.

The advancing machine learning techniques have been widely applied to data-driven dynamic stability assessment (DSA) in modern smart grids. However, how to extract critical spatial-temporal features from wide-area system stability dynamics still remains an open issue. Emphasizing on short-term voltage stability (SVS) assessment, this paper develops a novel sequential feature learning approach to address this problem in two steps. First, based on visualized voltage contours, it tactfully constructs a comprehensive spatial-temporal sequence model to dynamically characterize multiplex spatial-temporal SVS evolution trends. Second, the time series shapelet classification method is leveraged to subtly extract critical consecutive SVS features in sequential forms, i.e., the multidimensional shapelets (discriminative subshapes). Test results on the real-world Hong Kong power grid demonstrate the efficacy, adaptability, and scalability of the proposed approach for SVS assessment. In addition to the outstanding performances on online DSA, with its favorable interpretability, it is capable of providing intuitive insights into regional SVS patterns from spatial-temporal perspectives.

For power system dynamic security assessment (DSA), the conventional dynamic security region method is able to provide valuable information on security margins for preventive control. However, its event-based nature is likely to induce heavy computational burdens, especially in the presence of substantial presumed events. To tackle this challenging problem, this article develops an efficient time series data-driven scheme for batch DSA in a divide-and-conquer manner. First of all, with emphasis on short-term voltage stability, a novel u-shapelet (representative local trajectory)-based hierarchical clustering method is proposed to automatically divide various training cases into a handful of typical transient scenarios. Then, regressive shapelet learning is efficiently carried out to conquer individual scenarios, resulting in a group of high-precision security margin estimation models. With a desirable data-driven nature, the proposed scheme avoids time-consuming dynamic security region (DSR) characterization for each event, thereby achieving a significant speed-up for batch DSA. Test results on the realistic China Southern Power Grid illustrate its excellent performances on batch DSA.

In terms of machine learning-based power system dynamic stability assessment, it is feasible to collect learning data from massive synchrophasor measurements in practice. However, the fact that instability events rarely occur would lead to a challenging class imbalance problem. Besides, short-term feature extraction from scarce instability seems extremely difficult for conventional learning machines. Faced with such a dilemma, this paper develops a systematic imbalance learning machine for online short-term voltage stability assessment. A powerful time series shapelet (discriminative subsequence) classification method is embedded into the machine for sequential transient feature mining. A forecasting-based nonlinear synthetic minority oversampling technique is proposed to mitigate the distortion of class distribution. Cost-sensitive learning is employed to intensify bias toward those scarce yet valuable unstable cases. Furthermore, an incremental learning strategy is put forward for online monitoring, contributing to adaptability and reliability enhancement along with time. Simulation results on the Nordic test system illustrate the high performance of the proposed learning machine and of the assessment scheme.