The explosive growth of power load data has led to a substantial presence of abnormal data, which significantly reduce the accuracy of power system operation planning, load forecasting, and energy usage analysis. To address this issue, a novel improved GAN–Transformer model is proposed, leveraging the adversarial structure of the generator and discriminator in Generative Adversarial Networks (GANs). To provide the model with a suitable feature dataset, One-hot encoding is introduced to label different categories of abnormal power load data, enabling staged mapping and training of the model with the labeled dataset. Experimental results demonstrate that the proposed model accurately identifies and classifies mutation anomalies, sustained extreme anomalies, spike anomalies, and transient extreme anomalies. Furthermore, it outperforms traditional methods such as LSTM-NDT, Transformer, OmniAnomaly and MAD-GAN in Overall Accuracy, Average Accuracy, and Kappa coefficient, thereby validating the effectiveness and superiority of the proposed anomaly detection and classification method.
A coupled linear ultrasonic motor (LUSM) based on an eccentric constraint was proposed. Two pieces of oblique piezoelectric ceramics were arranged at each end of the elastomer, and the polarization direction of the ceramics was vertically upward. Using the tilting characteristics of the piezoelectric ceramics, the two ends of the fixed piezoelectric ceramics formed an eccentric restraint on the motor, providing conditions for the motor to generate coupled modes. When the elastomer of the motor generated the coupling vibration, the motion trajectories of the driving feet ends were oblique straight lines, and the oblique straight-line motion trajectories of the upper and lower driving feet ends were in opposite directions, driving the upper and lower sliders to run simultaneously. The stator parameters were optimized by using ANSYS to obtain larger amplitudes for the ends of the driving feet in both X and Z directions. The structure and operation principle of the motor are explained in detail. A prototype was fabricated to study the arrangement scheme with fixed constraints at the ends of the motor. The frequency–velocity characteristics, voltage–velocity characteristics, and mechanical characteristics of the motor were tested. The no-load speed and maximum output power were measured to be 45.9 mm/s and 3.24 mW.
Large-scale and widely distributed air-conditioning (AC) loads can be aggregated into load-type Virtual Power Plants (VPPs) to participate in peak-shaving ancillary services, thereby improving the allocation of demand-side electricity resources. However, current AC aggregation methods primarily focus on meeting peak-shaving instructions and generally employ fixed incentive pricing and proportional capacity allocation, making it difficult to balance user revenue and satisfaction and thereby constraining the flexibility of VPP demand-side regulation. This paper proposes a unified incentive-based demand response scheduling framework for both fixed- and variable-frequency AC loads across industrial, commercial, and residential scenarios. Based on the Equivalent Thermal Parameter model, AC loads are classified into curtailable and shiftable types, with their adjustable boundaries characterized by a Time-of-Use (TOU) elasticity-based interaction willingness model and a fuzzy load transfer rate model, respectively. A three-objective optimization model is established to maximize user revenue while minimizing user dissatisfaction and scheduling error, with incentive pricing and capacity allocation jointly optimized via Non-dominated Sorting Genetic Algorithm III (NSGA-III). Case studies are conducted on a load-type VPP covering three scenarios, namely a large industrial zone, a commercial zone, and a residential zone, under weekday and non-weekday TOU tariffs and three representative 1 h peak-shaving periods. Compared with a fixed-pricing benchmark, the proposed strategy increases total user revenue by 9.4% to 11.4% and reduces weighted average dissatisfaction by 0.27 to 1.92%. The case study results demonstrate that the proposed method can improve the trade-off between user revenue and satisfaction.
Distributed photovoltaic (PV) systems play an important role in supporting low-carbon and sustainable distribution network development; however, their high-penetration integration can cause nodal overvoltage, as midday PV generation often exceeds local demand and net reverse power flows accumulate along feeders. Existing studies mainly focus on technical mitigation, while entity-level responsibility quantification and its translation into differentiated mitigation tasks remain insufficiently explored. This paper proposes an overvoltage mitigation method integrating Shapley-value-based responsibility quantification with active–passive-responsibility-driven task allocation. Reverse power flow quotas are determined using the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) by considering the differentiated impacts of PV entities on power flow distribution and nodal voltage response. The load-weighted voltage deviation index is adopted as the coalition cost function, and the Shapley value is used to quantify each entity’s overvoltage responsibility share. Active and passive responsibilities are characterized through the net reverse power ratio and reactive power–voltage sensitivity, respectively, establishing a mapping mechanism from responsibility shares to active power curtailment and reactive power regulation tasks. A multi-objective optimization model that jointly considers line losses, PV curtailment, fairness deviation, and task-allocation deviation is constructed and solved using the whale optimization algorithm. Case studies on the IEEE 33-bus system show that the proposed method reduces the maximum network voltage from 1.25 pu to 1.052 pu and eliminates all overvoltage violations. Compared with a cluster-based scheme, the 24-h cumulative line loss is reduced by 30.6%, and the fairness deviation is significantly lowered, thereby supporting the sustainable, economical, and equitable operation of distribution networks with high distributed PV penetration.
Virtual power plants (VPPs) aggregate flexible resources, such as distributed photovoltaics (PV), energy storage, and flexible loads, to provide substantial reserve capacity for grid operation. However, the combined effects of renewable energy output uncertainty, load forecast errors, and user-bounded rationality responses lead to significant errors in traditional deterministic VPP reserve assessment methods, severely affecting the balance between system supply and demand. To address this challenge, this paper proposes a credible reserve assessment method that accounts for user-bounded rationality. First, thermodynamic models with on–off constraints for air conditioning loads, energy feasible region, and power constraint models for electric vehicles (EVs) and energy storage systems (ESSs), as well as PV forecast error models are established to characterize physical reserve boundaries. Second, prospect theory is introduced to describe user-bounded rationality and a logit-based response probability model is developed. Monte Carlo sampling and kernel density estimation are employed to derive credible reserve sets under different confidence levels, achieving a probabilistic quantification of VPP reserve capacity distribution. Case studies demonstrate that the proposed method accurately characterizes the probabilistic distribution characteristics of VPP reserve provision under multiple uncertainties, providing comprehensive and reliable assessment information for power dispatching agencies.
Lightweight day-ahead photovoltaic (PV) forecasting models encounter a significant technical challenge: under resource-constrained deployment conditions, it is difficult to simultaneously address weather-regime heterogeneity, maintain model interpretability, and preserve adaptability as operating conditions evolve. To address this issue, we propose a multimodal short-term photovoltaic (PV) forecasting method that integrates weather-mode partitioning using the Clustering Using Representatives (CURE) algorithm with a self-updating Batch-Lasso model. First, the meteorological-PV dataset is partitioned along two dimensions by combining seasonal grouping with CURE clustering within each season, producing representative weather modes and enhancing the fidelity of weather pattern classification. Second, to extract informative predictors from high-dimensional meteorological inputs while maintaining interpretability, we formulate per-mode Lasso regression and adopt the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA) to efficiently solve for the sparse regression coefficients. Third, we introduce a batch-based self-update and correction mechanism with rollback verification, enabling the mode-specific models to be refreshed as new historical data become available while preventing performance degradation. Compared with representative machine learning baselines, the proposed method maintains competitive accuracy with substantially lower computational and storage overhead, enabling high-frequency and energy-efficient inference on resource-constrained terminals, thereby reducing operational burdens and computational energy costs and better meeting the deployment needs of sustainable energy systems under heterogeneous weather conditions.