Existing methods for power load anomaly detection suffer from several limitations, including insufficient extraction of multi-scale temporal features, difficulty in capturing long-range dependencies, and inefficient fusion of heterogeneous spatiotemporal information. To address these issues, this study proposes the TGCformer, an enhanced Transformer-based model designed for dynamic spatiotemporal feature fusion. First, a dual-path spatiotemporal feature extraction module is constructed. The temporal path utilizes TSFresh to enhance the explicit pattern representation of the load sequences, while the spatial path employs an improved GATv2 to model dynamic correlations among grid nodes. Together, these two paths provide more interpretable and structured inputs for the Transformer encoder. Subsequently, a multi-head cross-attention mechanism is designed, where temporal features serve as the Query and graph embeddings as the Key and Value, to guide the feature fusion process. This design ensures the effective integration of complementary information while suppressing noise. Experimental results on the public Irish dataset demonstrate the effectiveness of the proposed model. Specifically, TGCformer achieves average F1-score improvements of 0.35 and 0.53 compared with InceptionTime and XceptionTime, respectively.

Photovoltaic (PV) generation is highly intermittent and weather-sensitive, which increases the difficulty of short-term power forecasting, especially under extreme weather conditions. To address this issue, this paper proposes a Bayes-BiLSTM-based method for short-term PV power forecasting under extreme weather conditions. According to the dynamic characteristics of PV generation, multiple meteorological factors are analyzed for feature selection and sample extraction, and a Bayes-BiLSTM model is constructed whose hyperparameters are adaptively tuned by Bayesian optimization. To further improve generalization under severe output fluctuations, an expert classification framework is introduced together with convolutional neural networks (CNNs), an adaptive Huber loss function, and a gating-based fusion mechanism. Experimental results show that the proposed method significantly reduces forecasting errors under extreme weather, with the RMSE decreasing from 4.0357 to 1.7192, the MAE decreasing from 2.3560 to 1.1868, and the $\mathrm{R}^{2}$ increasing from 0.2061 to 0.8035. The proposed method therefore improves both forecasting accuracy and robustness and provides useful support for grid dispatch and renewable-energy management.

With the continuous growth of electric vehicle deployment, the temporal and spatial aggregation of charging demand has become increasingly significant. The mismatch between vehicles and charging facilities is becoming more prominent, posing new challenges to user charging experience and distribution network operation. Charging guidance involves not only grid side operating conditions, but also traffic side factors such as detour, queuing, and route choice. As a result, user charging decisions exhibit clear multi source coupling characteristics. Therefore, how to incorporate user loss arising from the transportation system into a unified electrical modeling framework has become a key issue in charging guidance research. To address the difficulty of existing methods in jointly representing multi source factors from the road network side and the grid side, this paper proposes a virtual impedance based charging guidance method for electric vehicles. First, a virtual impedance model composed of road network cost and grid cost is constructed to provide a unified representation of the overall charging cost. Then, a charging guidance model is established by incorporating heterogeneous user preferences. Finally, a two-layer charging guidance strategy is designed, and case studies are conducted to verify the effectiveness of the proposed method. The results show that the proposed method can improve the rationality of user charging decisions, relieve local distribution network pressure, and enhance the overall guidance performance.

Affected by the large-scale renewable energy consumption targets, the role of the power load side has shifted from the traditional energy consumer to a complex prosumer. The nodal price mechanism in the electricity market further demands that electricity users possess flexibility to cope with the challenges and opportunities arising from complex conditions such as policies, environment, weather, etc. Industrial users, as an important category of electricity consumers, urgently need to develop quantitative methods for constructing industrial virtual power plants (IVPPs) to effectively participate in the grid interactions. Based on this, this paper proposes a quantitative assessment approach and optimization model for IVPP applications, for the purpose of precise implementation and practical regulation of industrial flexibility utilizations. Firstly, a multi-objective optimal scheduling model for IVPPs is established by considering the complex and various environments on load side. Then, in order to improve the performance of industrial load resources, a multi-level refinement index system is proposed to quantitatively evaluate the IVPP’s regulation potentials from the perspective of aggregation level. Lastly, simulation results show that the proposed model is sufficiently stable and accurate to support the industrial flexibility utilization. The quantitative IVPP model provides a paradigm for flexibility interactions of industrial loads, and further serves as the technical foundation for transforming their role from passive to active in the electricity market.

With the rapid development of distributed renewable energy, demand response (DR) has emerged as a crucial approach to maintaining the balance between electricity supply and demand and facilitating the integration of renewable energy sources. However, effective implementation of DR relies on fine-grained visibility into user load characteristics. In parallel, non-intrusive load monitoring (NILM) has received increasing research attention in recent years, driven by the growing emphasis on energy efficiency and the rise of intelligent home energy management. NILM technology, by disaggregating user-level electrical loads into individual appliances, provides detailed consumption information that can be leveraged to design refined and personalized strategies essential for effective DR implementation. This paper presents an integrative review of the collaborative optimization in DR empowered by NILM and explores the major technical challenges and corresponding solutions in this field. The review identifies key technical challenges, such as the trade-off between real-time performance and accuracy, privacy preservation, and scenario adaptability, and discusses corresponding design guidelines. Key research findings reveal that while deep learning and lightweight models enhance real-time responsiveness, the effective integration of NILM into practical DR systems depends not only on balancing centralized coordination efficiency with privacy preservation and ensuring the cross-scenario adaptability across residential, commercial, and industrial settings, but also on evaluating these improvements in relation to the specific stages of DR implementation. By establishing a structured understanding of NILM-enabled DR, this paper paves the way for the scalable, efficient, and intelligent deployment of DR in future power systems with deep renewable integration.

Incomplete measurements, topology uncertainty, and abrupt reconfigurations significantly degrade situational awareness in modern electric distribution systems, especially under high renewable penetration. Conventional sequential 'identify-then-estimate' schemes suffer from severe error propagation and delayed responsiveness. To bridge this critical gap, this work proposes a physics-informed graph-masked autoencoder (PI-GMAE) framework, integrating topology identification (TI) and state estimation (SE) into a unified, self-supervised paradigm. Numerical validation confirms that PI-GMAE achieves instantaneous topology tracking and accurate state recovery even under 50% measurement masking, effectively neutralising the error accumulation of sequential methods to support real-time situational awareness in modern distribution systems.

To address the instability of islanding detection performance in AC V2G bidirectional charging and discharging systems under varying grid strength conditions, this paper proposes an SCR-aware grid-strength adaptive AFDPF islanding detection method. The effective short-circuit ratio ($S C R_{\text {eff }}$) is introduced to achieve real-time quantification of grid strength, based on which a mapping relationship between grid strength and control parameters is established to adaptively adjust disturbance amplitude, positive feedback gain, and detection thresholds. Furthermore, a two-stage decision mechanism incorporating a direction consistency constraint is designed to enhance detection reliability and suppress false triggering. Simulation studies are carried out under weak, medium, and strong grid conditions using the MATLAB/Simulink platform. The results demonstrate that the proposed method can reliably detect islanding over a wide range of grid strengths. The detection speed increases with grid strength, while frequency fluctuations are effectively suppressed under weak grid conditions. Compared with conventional fixed-parameter AFDPF methods, the proposed strategy significantly reduces the nondetection zone (NDZ), improves detection speed, and enhances system robustness, showing strong potential for practical applications in AC V2G systems.

With high renewable penetration, electricity prices are volatile, multi-scale, and nonstationary. Fixed-basis time-frequency methods limit adaptivity, two-dimensional reshaping oversmooths spikes and shifts phase, and attention models become costly as horizons grow. To address these issues, we propose TimesWave, an end-to-end architecture for multivariate long-sequence forecasting. The method employs learnable wavelet lifting as a time-frequency front end and a period-aware dual-core backbone for representation learning. Adaptive lifting produces a cleaner low-frequency approximation and structured detail components. The approximation is then mapped to a two-dimensional periodic grid, where a multi-scale convolutional branch captures local patterns and a periodic attention branch models cross-period dependencies. The final forecast is obtained by reconstructing through inverse lifting with shared weights. Across two real-word market datasets and a range of horizons, TimesWave delivers higher accuracy and more stable long-horizon performance than strong baselines. Ablation studies further show that removing the adaptive lifting or either branch of the dual core leads to significant degradation, confirming their complementarity. These findings indicate that tightly coupling adaptive time-frequency analysis with period-aware feature extraction provides reliable price forecasts for risk-sensitive electricity markets.

Accurate load forecasting across heterogeneous temporal scales is fundamental for secure and efficient grid operations. Due to the distinct characteristics of forecasting tasks at varying time scales, existing research has predominantly focused on single-scale models, limiting the exploration of practical, unified multi-scale forecasting frameworks and underutilizing cross-scale dependencies. To address these gaps, this study proposes a unified multi-scale framework for one-step forecasting across multiple resolutions. A scale-aware shared Transformer encoder with learnable scale embeddings is introduced, offering a unified parameterization that facilitates effective cross-scale information exchange. Resolution-specific branches are employed to align model capacity with the statistical properties of each scale. Additionally, a resolution-adaptive training objective is formulated, utilizing scale-specific loss functions with tunable weights to balance robustness, convergence, and inter-task optimization. Evaluations on real-world load data across three temporal resolutions demonstrate the framework's capability to deliver comparatively accurate forecasts across scales.

In modern smart grids, accurate and synchronized time signals are essential for effective monitoring, protection, and control. Various time synchronization methods exist, each tailored to specific application needs. Widely adopted solutions, such as GPS, however, are vulnerable to challenges such as signal loss and cyber-attacks, underscoring the need for reliable backup or supplementary solutions. This paper examines the timing requirements across different power grid applications and provides a comprehensive review of available time synchronization mechanisms. Through a comparative analysis of timing methods based on accuracy, flexibility, reliability, and security, this study offers insights to guide the selection of optimal solutions for seamless grid integration.

Accurate forecasting of photovoltaic (PV) power is crucial for real-time grid balancing and storage system optimization. However, due to the intermittent and fluctuating nature of PV power generation, achieving accurate PV power forecasting remains a challenge. In this paper, we propose a novel approach for multi-site intra-hour PV power forecasting. Different from current work which predicts the power of each PV station independently, we predict the power of each PV station simultaneously by considering the inherent spatio-temporal correlation with other PV stations and develop a novel graph network named DEST-GNN. In DEST-GNN, an undirected graph is used to represent the dependence of these PV stations. Each PV station is represented by a node and the spatio-temporal correlation of any two PV stations is represented by an edge between them. To improve the accuracy of prediction, sparse spatio-temporal attention is adopted to filter out the weak associations of these PV stations. We then develop an adaptive graph convolution network (GCN) that adopts an adaptive adjacency matrix and a temporal convolution network to capture the hidden spatio-temporal dependency of these PV stations. Experimental studies using datasets from Alabama and California, provided by the National Renewable Energy Laboratory (NREL), demonstrate the effectiveness of DEST-GNN. For the Alabama dataset, DEST-GNN achieves a mean absolute error (MAE) of 0.49 over a 12-mon training scale. Furthermore, DEST-GNN attains an MAE of 0.42 on the California dataset, continuing to exhibit its strong forecasting capabilities.

As Medium Voltage Direct Current (MVDC) technology becomes prevalent in modern all-electric ships, the complexity of shipboard power systems (SPS) is increasing. The dynamic interactions among pulsed loads, battery energy storage systems (BESS), and diesel generators pose challenges to grid stability, making rapid and accurate fault diagnosis critical for ensuring safe vessel operation. Traditional fault diagnosis methods face limitations in adapting to the dynamic conditions and complex fault modes of shipboard systems. To address this issue, this paper proposes a deep learning model based on a Gated Recurrent Unit (GRU) for fault diagnosis in shipboard microgrids. This approach leverages the inherent advantages of GRU networks in processing time-series data to directly learn and extract dynamic fault features from voltage signals, distinguishing between normal and faulty operating states. Validated on a detailed MVDC shipboard system simulation, the proposed GRU model demonstrates significant potential in terms of diagnostic accuracy and reliability, offering robust technical support for an efficient and intelligent SPS monitoring system.

Abnormal power usage detection is vital for power system security and efficient energy use. Existing methods are hampered by low sampling rates, insufficient feature extraction, and limited accuracy, which impede cost effective large scale deployment and raise privacy concerns. This paper proposes a low frequency electrical characteristic based abnormal power usage detection scheme. First, an abnormal power usage detection model is constructed by coupling active power and power factor under low frequency measurement conditions of existing smart meters. Second, the scheme embraces non-intrusive load monitoring (NILM) to safeguard user privacy and markedly reduce computational burden. Validation based on smart meter data indicates that the proposed scheme offers a more efficient, simpler, and more cost-effective solution for deploying advanced NILM-based metering systems in residential settings.

Non-intrusive load monitoring (NILM) can infer load profiles for each individual appliance from aggregated power consumption signals without installing extra sub-meters. However, performance of traditional energy disaggregation methods deteriorates in complex environments, especially susceptible to the presence of other high power consumption appliances. Practicalities are also limited by diversity of household load patterns and measurement errors. In order to address these problems, a hybrid deep learning model consisting of two steps is proposed in this paper. First, an improved variational autoencoder (VAE) structure is introduced for preliminary energy disaggregation, where the encoder and decoder layers are long short-term networks (LSTM) to extract temporal characteristics of active power signals. Afterward, a post-processing method based on Siamese one-dimensional convolutional neural network (S-1D-CNN) is adopted to remove incorrectly predicted activation segments of target appliances. Experiments are conducted on two public datasets, and results show remarkable improvements on prediction accuracy over other deep learning methods. Both transferability and stability of the proposed model are verified under different working conditions.

ABSTRACT With the gradual liberalization of the carbon market and distributed trading market, the economic incentive trading market mechanism has become an effective way to promote carbon emission reduction in microgrids. At present, most of the existing studies on the low‐carbon operation of integrated energy systems focus on the source side and rarely extend to the load side, and do not consider the demand response characteristics of different loads. Therefore, based on the carbon flow tracing method of the power system, this paper presents a model to adjust the load side operating state of the power system by using price incentive. Firstly, the carbon flow tracing model of the integrated energy system is established, and carbon flow indexes such as node carbon potential are obtained. At the same time, considering different load types, the carbon reduction response mechanism of two loads is established through carbon trading. On this basis, according to the carbon flow index, the two‐stage optimal scheduling model of the power network with the coordination and interaction between the two sides of the source and load is established and solved. The simulation results show that the model combines carbon trading and demand response, which can effectively reduce carbon emissions and significantly improve the environmental benefits of the system.

Accurate and rapid indoor positioning is essential for realizing the Internet of Things. Although satellite-based positioning works well outdoors, it often fails indoors due to signal blockage. Radio Frequency Identification (RFID) positioning technology has attracted attention due to its low cost, ease of deployment, and contactless operation. However, traditional methods based on Received Signal Strength Indicator (RSSI) are vulnerable to environmental interference, reducing accuracy. This paper proposes a passive RFID indoor positioning method that uses multi-frequency phase differences ranging. A Gauss-Kalman filter is applied to reduce the influence of environmental noise on phase data. A multi-frequency phase difference distance model is established, and the Chinese Remainder Theorem with a closed-form solution is used to solve phase ambiguity. Finally, a weighted Levenberg-Marquardt algorithm refines the target position, improving convergence and accuracy. Experiments conducted in different indoor environments demonstrate that the proposed method achieves an average positioning error of 0.38 m and a maximum error of 0.5 m, meeting the requirements of indoor positioning applications.

Amid the energy transition, power grids feature a high proportion of new energy and power electronics, posing challenges to traditional power supply capacity evaluation, such as complex cross-section coupling, strong DC stability requirements, and difficulties in multi-objective coordination. Existing methods over-rely on the N-1 constraint, overlook cross-section coupling and DC support, and struggle to balance power supply, safety, and reserve in single-objective optimization. This paper proposes an intelligent multi-objective mixed-integer programming evaluation method, constructing a three-objective model encompassing power supply, cross-section margin, and reserve balance. It systematically incorporates cross-section coupling and employs the NNC-AHP-TOPSIS approach to obtain the Pareto frontier and make decisions. Case verification shows that the method has significant advantages in accuracy and efficiency, providing theoretical and tool support for power grid dispatching and planning.

With the development of medium-voltage direct current (MVDC) technology, MVDC shipboard power system has begun to be widely applied. Adopting reasonable and economical means to effectively monitor the operating status of shipboard power systems, thereby promptly identifying and preventing fault states of shipboard loads, is an important way to improve the safety and reliability of ship operations. The shipboard's operating status and spatial conditions greatly restrict the direct acquisition of data on the operating status of each load. To address this issue, firstly this paper uses MATLAB/Simulink to construct the topology of the MVDC shipboard power system and conducts mathematical modeling for the required loads. Secondly, this paper runs simulations, collects electricity consumption data at the bus of the shipboard power system under different load states, and builds datasets. Finally, this paper uses the K-Means clustering algorithm to cluster the unlabeled dataset, identifies and explores load fault states, and selects corresponding evaluation indicators to verify the effect and feasibility of the clustering algorithm, providing a reliable solution for efficiently and accurately identifying and early warning of load fault states in MVDC shipboard power systems.