Huifang Wang

Deep learning methods for the power quality disturbance (PQD) classification encounter significant deployment challenges in power system applications. With the increasing integration of renewable energy, electric vehicles, and other non-traditional loads, the out-of-distribution (OOD) data occurs frequently due to evolving power system conditions, which represents a critical yet underexplored challenge. In this study, the OOD problem in the PQD classification is formally defined, where models encounter the absence of disturbance patterns during training. Leveraging the unique characteristics of PQD data, a four-stage multimodal OOD detection method is developed for the OOD-driven framework to enhance PQD classification. The proposed method first transforms PQD waveforms into dual-channel Gramian Angular Field (GAF) maps, then extracts discriminative features with vision Transformers. Mahalanobis distance is employed for OOD scoring, and a multimodal auxiliary criterion is introduced to enhance the distinction between in-distribution (IND) and OOD. Comprehensive case studies are conducted using both simulated IEEE standard-compliant waveforms and practical PQD data. Results applied to the simulated PQD data demonstrate that the proposed method can achieve a 2.10% of false positive rate at 95% true positive recall (FPR95), outperforming all previous methods. Practical applications demonstrate a remarkable 9.64% improvement in the PQD classification accuracy compared to the unfiltered dataset containing OOD samples.

Multiple energy storage systems (ESSs) often face imbalances in charging–discharging operations, as well as the uncertainties of practical scenarios and influencing factors. To address these challenges, this study proposes a user‐side cloud energy storage (CES) model with active participation of the operator. This CES model incorporates adjustable time‐of‐use (TOU) electricity pricing and state‐of‐charge (SOC) management. In the configuration process, the net load scenario generation reduction is performed first. Subsequently, demand response is implemented based on the updated TOU pricing. To address the imbalance of ESSs, an improved multiobjective particle swarm optimization is employed, followed by access verification of the multi‐ESS aggregation. In the dispatch process, a two‐stage interval optimization model is adopted. Specifically, day‐ahead scheduling determines the SOC limit interval, and intra‐day scheduling achieves rolling optimization to determine the exact charging–discharging duration. This ensures that the charging–discharging cycles are controllable, orderly, and efficient. Ultimately, a fair settlement method based on optimal pricing of various fees within the “cloud” is proposed, ensuring sustainable revenue growth for all types of users. A case study demonstrates that the proposed methods can achieve multifaceted value in energy management and enhance the socioeconomics of user‐side ESS projects.

The increasing integration of Inverter Interfaced Distributed Generators (IIDGs) into distribution networks has led to a more active nature of the grid. However, the accuracy-speed trade-off caused by iterative algorithms and simplified models in the short-circuit current calculation method for active distribution networks (ADNs) is becoming more prominent. To address this issue, we propose a novel approach that utilizes Graph Convolutional Neural Networks (GCNs) for short-circuit current calculation in ADNs. Our study explores the characteristics of short-circuit current in different network structures and evaluates the feasibility of representing electrical quantities using a graph data format. The proposed method employs the GCN model to calculate multi-output ADN short-circuit current and investigates the block construction of the GCN model. Algorithm analysis demonstrates that our method effectively calculates network-wide short-circuit current under various network structures, IIDG penetration rates, and fault conditions, meeting both accuracy and computational speed requirements. The proposed method offers several advantages, including superior precision, rapid computation, minimal hardware resource utilization, and robust resistance to interference.

In simulations of three-dimensional transient physics filled through a numerical approach, the order of the equation set of high-fidelity models is extremely high. To eliminate the large dimension of equations, a model order reduction (MOR) technique is introduced. In the existing MOR methods, the block Arnoldi algorithm-based MOR method is numerically stable, achieving a passively reduced order model. Nevertheless, this method performs poorly when it is applied to very wide-frequency transients. To eliminate this deficiency, multipoint MOR methods are emerging. However, it is hard to directly apply an existing multipoint MOR method to a 3-D transient field equation set. The implementation issues in a reduction process (such as the selection of expansion points, the number of moments matched at a point and the error bound) have not been explored in detail. In this respect, an adaptive multipoint model reduction model based on the Arnoldi algorithm is proposed to obtain the reduced-order models of a 3-D temperature field. The originality of this study is the proposal of a novel adaptive algorithm for selecting expansion points, matching moments automatically, using a posterior-error estimator based on temperature response coupled with a network topological method (NTM). The computational efficiency and accuracy of the proposed method are evaluated by the numerical results from solving the temperature field of a prototype insulated-gate bipolar transistor (IGBT).

The widely applied Structure-Preserving Reduced-order Interconnect Macromodeling (SPRIM) algorithm cannot retain the reciprocity or the block structure of the circuit matrices of the original system, being inherent to RCL circuits; and consequently it is harder to synthesize a Passive Reduced-order Interconnect Macromodeling Algorithms (PRIMA) model into a practical circuit. In order to overcome this deficiency, this paper proposes an improved SPRIM algorithm to maintain the structure features of the input-output correlation matrix. Firstly, the algorithm in this paper reorders the modified nodal analysis (MNA) equations. An Input-Output structure Preserving Order Reduction (IOPOR) technique applicable to SPRIM is then proposed. Also, one combines the IOPOR technique with the conventional SPRIM algorithm to obtain a new projection matrix that can maintain the input and output correlation matrix of the original system. The feasibility and effectiveness of the proposed algorithm are verified through numerical analysis and calculation of typical problems.

Full-duplex communication technology can make full use of spectrum resources through simultaneous co-frequency transmission, but the interference of the signal sent by the device itself degrades performance. The paper proposes a joint antenna and relay selection algorithm based on Q-learning, which selects the best relay with the best transmit-receive configuration for the communication between the source and the target. Specifically, by learning the dual-medium cooperative communication environment to obtain feedback in terms of interruption probability and channel capacity, so as to determine the best relay and antenna. Finally, the paper verifies the effectiveness of the algorithm by simulation.

Timely and accurate fault location techniques can reduce power outages and improve power supply reliability. This paper proposes a new fault location method for active distribution networks (ADNs) which utilizes limited synchronized measurements (PMUs). The proposed method formulates the fault location problem as the estimation of a block-sparse fault injection current signal which can inherently reveal fault types and locations. A modified block-sparse Bayesian learning (BSBL) algorithm is utilized to estimate this block-sparse signal by exploiting its both block structure and intra-block correlation. In addition, the proposed method provides a strategy for the optimal placement of much fewer PMUs that will not yield a complete observable system. The proposed method is able to pinpoint the fault position accurately, and is applicable to both balanced and unbalanced ADNs. The effectiveness of the proposed method is examined on the IEEE 123-node distribution networks with DGs and the results are discussed thoroughly.

In the transient analysis of an engineering power electronics device, the order of its equivalent circuit model is excessive large. To eliminate this issue, some model order reduction (MOR) methods are proposed in the literature. Compared to other MOR methods, the structure-preserving reduced-order interconnect macromodeling (SPRIM) based on Krylov subspaces will achieve a higher reduction radio and precision for large multi-port Resistor-Capacitor-Inductor (RCL) circuits. However, for very wide band frequency transients, the performance of a Krylov subspace-based MOR method is not satisfactory. Moreover, the selection of the expansion point in this method has not been comprehensively studied in the literature. From this point of view, a broadband enhanced structure-preserving reduced-order interconnect macromodeling (SPRIM) method is proposed to reduce the order of equation sets of a transient interconnect circuit model. In addition, a method is introduced to determine the optimal expansion point at each frequency in the proposed method. The proposed method is validated by the numerical results on a transient problem of an insulated-gate bipolar transistor (IGBT)-based inverter busbar under different exciting conditions.