With the increasing prevalence and evolvement of ongoing technologies in household smart meters, nonintrusive load monitoring (NILM) becomes a convenient and cost-effective solution for appliance-level energy monitoring and analysis. As a vital tool for more informed electricity usage management, NILM is naturally intersected with the emerging field of deep learning. However, the predominant focus in deep learning-based NILM research is on enhancing the structure of neural networks, giving little emphasize on the data quality problem which heavily influences the accuracy and robustness of NILM. To address this issue, this article analyses the point-type and pattern-type abnormal data that may affect the accuracy of NILM, and proposes a data anomaly detection method accordingly based on an enhanced cycle-consistent generative adversarial network, which can represent the characteristics of load series in a low-dimensional latent space. Furthermore, by this latent space, we present an anomaly calibration approach based on clustering and nearest neighbors approximation. Extensive experiments using the open data sets are designed to demonstrate the effectiveness of the proposed method in improving NILM accuracy in both on/off states identification and appliance-level load prediction.

Uncertainty sets (USs) based on historical data have been applied for accurately characterizing the uncertainty of renewable energy resource (RES) unit outputs in robust energy scheduling involving virtual power plants (VPPs). However, it remains highly challenging to develop scheduling solutions that optimally balance between security and economic efficiency and the lowest computational burden. This involves constructing the smallest possible linear-form US that encompasses RES uncertainty data with a minimum number of vertices. The present work addresses these challenges by developing a data-driven minimum-volume ellipsoid US (EUS) with flexible confidence levels. The number of vertices in the obtained EUS is reduced to improve the computational efficiency of the solution process by approximating the EUS using a hybrid polyhedron US (HPUS) composed of rectangular and diamond USs. Finally, a vertex-based column-and-constraint generation algorithm, which can avoid falling into locally optimal solutions, is designed to solve the robust VPP energy scheduling model with the HPUS. The effectiveness and superiority of the proposed US approach and algorithm are verified based on a practical VPP system in South China.

Hybrid ac/dc microgrids (HMGs) have emerged as a promising paradigm for integrating large numbers of inherently uncertain and correlated renewable energy sources (RESs). To address the uncertainty introduced by extensive RES integration, a quad-level energy management model is proposed for HMGs, incorporating a novel data-driven uncertainty set. Specifically, a pair convex hull uncertainty set (PCHUS) is developed with adjustable size, which utilizes a graph neural network to identify and exclude outliers in RES data. This approach provides trustworthy RES data at any given confidence level. Then, a quad-level energy management model is designed to determine the minimal-size PCHUS among all possible options, ensuring the least conservative solution, while maintaining robustness against RES fluctuations. Furthermore, a modified version of Taguchi’s orthogonal array testing (TOAT) method, termed quasi-TOAT, enhances the proposed solution algorithm. This modification enables parallel processing capabilities, significantly improving both the global optimum-seeking process and computational efficiency. To validate the proposed approach, the proposed uncertainty set and enhanced algorithm are compared against existing methods using a practical HMG case study. The results demonstrate the effectiveness and superiority of the proposed methodology in managing uncertainty within HMGs.

Due to the potential devastating impact on modern Internet-of-Things (IoT) integrated power grids, thefalse data injection attack (FDIA) has become a major concern. This article proposes an FDIA approach against state estimation without the knowledge of system parameters considering measurement noise. The proposed approach is able to mitigate the impact of measurement noise by utilizing the low-rank characteristic of the measurement data matrix, and can recover partial singular vectors of the state estimation Jacobian matrix (SEJM), based on which an unobservable attack can be launched. Besides, the scenario that only partial sensors can be tampered is investigated, and a matrix extension or split strategy is used to modify the matrix size, which makes the proposed method can be applied into the power grid of arbitrary scale. The presented method is capable of achieving a higher attack successful rate as well as requiring less amount of measurement data. Numerous cases demonstrate the effectiveness and advantages of the proposed method over other FDIA approaches in the scenario system parameters are unavailable.

False data injection attacks (FDIAs) represent a significant threat to power grid cybersecurity, designed to manipulate crucial measurement data and thereby compromise the operation of power grids. This article proposes an ac FDIA method based on tensor principle component analysis (TPCA), requiring no prior knowledge of system parameters. The goal of the proposed approach is to produce false data that can break through the bad data detection (BDD) of realistic ac state estimation. Specifically, ac state estimation model is transformed into a tensor representation, encapsulating measurement variables, state variables, and system parameters as a combination of multiple tensor products. Following this, by formulating multiple measurement data into a tensor, TPCA is used to decompose the measurement data tensor to obtain a space of matrices. Subsequently, the vector of false data ensuring the stealthiness is produced by finding a rank-1 approximation of matrices in this space. Notably, the proposed method distinguishes itself from existing parameter-free FDIA methods by eschewing any simplification or approximation of ac state estimation model. Numerous cases in IEEE 5, 14, 57, 118, 300-bus, European 1354-bus, and Polish 3120-bus testing systems provide substantial evidence that the proposed approach can obtain higher attack successful rate. It achieves 99.3% attack successful rate on average against the common$\chi ^{2}$BDD with 0.9 confidence level. And compared with existing methods, the attack successful rate improves 4%, 2%, 13%, 11%, 10%, and 27% in these six systems, respectively.

Remaining useful life (RUL) prediction is of fundamental importance in reliability analysis and health diagnosis of complex industrial systems. Aiming at improving the prediction accuracy, this article proposes a normalizing flow embedded sequence-to-sequence (seq2seq) learning method to predict the RUL of an asset or a system. This method introduces a block of normalizing flow into the middle area of the familiar encoder–decoder structure of the seq2seq model. This normalizing flow enjoys the remarkable representation ability for the nonlinearity between input sequential data and outputs and enables the original seq2seq model to be more suitable for vibration signals of engines. The encoder and the decoder, which fall before and after the normalizing flow, are both built by gated recurrent units. Besides, a one-hot coding of clustering is concatenated with measurement data to indicate the frequently shifting vibration state, and a sensor selection method is designed to drop some weakly related and ineffective variables. Our method is tested and further analyzed by 2008 IEEE PHM challenge data (PHM08), of which many practical preprocessing methods are conducted. Numerous tests verify that our method outperforms other related deep learning methods for RUL estimation.