国家电网

Multiview multilabel learning with incomplete views and missing labels (MVMLL-IVML) aims to make multilabel prediction by exploring incomplete information from multiview multilabel data. Existing methods learn a unified view weight to fuse view information based on the assumption that the view weights for the same view across all samples are consistent. However, in practical applications, the view weights for different samples may be different, even the difference is significant. For the drawbacks, we propose dual-weight fusion framework (DWFF). First, we employ autoencoders to obtain high-level features (HLF) for each view and introduce contrastive learning to increase the ability of the autoencoder to capture rich-view semantic information. Second, we propose the dual-feature comparison (DFC) strategy to obtain view weights by comparing the original features with the reconstructed features. Third, all features are fused based on the obtained view weights and the missing view indicator matrix. Finally, the fused HLF are fed into the classifier. Extensive experimental results demonstrate the effectiveness of the proposed DWFF.

Referring Video Object Segmentation (RVOS) requires segmenting specific objects in a video guided by a natural language description. The core challenge of RVOS is to anchor abstract linguistic concepts onto a specific set of pixels and continuously segment them through the complex dynamics of a video. Faced with this difficulty, prior work has often decomposed the task into a pragmatic `locate-then-segment' pipeline. However, this cascaded design creates an information bottleneck by simplifying semantics into coarse geometric prompts (e.g, point), and struggles to maintain temporal consistency as the segmenting process is often decoupled from the initial language grounding. To overcome these fundamental limitations, we propose FlowRVS, a novel framework that reconceptualizes RVOS as a conditional continuous flow problem. This allows us to harness the inherent strengths of pretrained T2V models, fine-grained pixel control, text-video semantic alignment, and temporal coherence. Instead of conventional generating from noise to mask or directly predicting mask, we reformulate the task by learning a direct, language-guided deformation from a video's holistic representation to its target mask. Our one-stage, generative approach achieves new state-of-the-art results across all major RVOS benchmarks. Specifically, achieving a J&F of 51.1 in MeViS (+1.6 over prior SOTA) and 73.3 in the zero shot Ref-DAVIS17 (+2.7), demonstrating the significant potential of modeling video understanding tasks as continuous deformation processes.

Denoising-based diffusion transformers, despite their strong generation performance, suffer from inefficient training convergence. Existing methods addressing this issue, such as REPA (relying on external representation encoders) or SRA (requiring dual-model setups), inevitably incur heavy computational overhead during training due to external dependencies. To tackle these challenges, this paper proposes VAE-REPA, a lightweight intrinsic guidance framework for efficient diffusion training. VAE-REPA leverages off-the-shelf pre-trained Variational Autoencoder (VAE) features: their reconstruction property ensures inherent encoding of visual priors like rich texture details, structural patterns, and basic semantic information. Specifically, VAE-REPA aligns the intermediate latent features of diffusion transformers with VAE features via a lightweight projection layer, supervised by a feature alignment loss. This design accelerates training without extra representation encoders or dual-model maintenance, resulting in a simple yet effective pipeline. Extensive experiments demonstrate that VAE-REPA improves both generation quality and training convergence speed compared to vanilla diffusion transformers, matches or outperforms state-of-the-art acceleration methods, and incurs merely 4% extra GFLOPs with zero additional cost for external guidance models.

This paper proposes a Voltage-Dynamic-Distributionally Robust Co-Planning Framework (VDD-CPF) for the optimal siting and sizing of grid-scale energy storage systems (ESSs) in distribution networks with high levels of renewable uncertainty. The framework employs a second-order cone AC power-flow model, together with the Voltage Collapse Proximity Index (VCPI), which incorporates a Dynamic Thermal Rating (DTR) model to capture weather-variable variations in line ampacity and also accounts for stability margins. Renewable load uncertainties are formulated as distributionally robust chance constraints; thus, the planning decision on ESS is subject to frequent extreme events. Energy storage system (ESS) operation is modeled to account for state-of-charge variations, the ratio of the battery’s charge or discharge (C-rate) limits for demand and supply, and throughput-based degradation to maintain realistic long-term asset behavior. A spatial column-and-bar design is carried out for progressive allocation sites and examination of operational feasibility under multiple scenarios. The methodology is demonstrated through the IEEE-33 bus system using realistic PV and true hour power profiles. The simulation outputs of the VDD-CPF site selection and sizing of ESS show that the distribution operation resilience in terms of the voltage flexibility, thermal rating, and dynamic stability margins have been improved by the methodological inclusion. The findings prompt the need for various approaches to reflect the dynamics of the stability indices, thermal behavior, and uncertainty. It is believed that the methodology will be a useful tool for the short yield and long-dated investment planning.

New-Type Power Systems (NTPS) are emerging as a response to deep decarbonization goals, driven by high shares of renewable energy, extensive electrification, and advanced digital control. These systems are designed to be clean, flexible, intelligent, and interactive. However, integrating large numbers of Renewable Energy Sources (RESs) and Electric Vehicles (EVs), especially under EV flexible load regulation operation, introduces significant challenges related to uncertainty, system coordination, and scheduling complexity. Therefore, this paper presents a novel two-stage sequential scheduling framework that jointly schedules coordinates Generation-Grid-Load-Storage (GGLS) resources within an NTPS context. The first stage establishes an economic dispatch model that simultaneously optimizes the operation of Distributed Generators (DGs), grid electricity purchases, and Interruptible Loads (ILs), while considering market signals, forecasted prices, and operating constraints. The second stage formulates a reactive power optimization model to reduce network losses and enhance voltage stability by adjusting DGs and reactive power devices, using the first-stage schedules as fixed inputs (i.e., without iterative feedback to update first-stage decisions). To solve the mixed-variable scheduling problem efficiently, a modified Feedback Wolf Pack Algorithm (FWPA) is developed. The algorithm introduces dynamic feedback control, allowing for real-integer variable decoupling, search space dimensionality reduction, and improved global convergence. Simulation results on the IEEE 33-bus system validate the proposed method. For the modified IEEE 33-bus active distribution network, the proposed framework limits peak-load active power losses to about 42.4 kW at a 5.02 MW demand and converges in approximately 72 s, yielding around 50 % and 75 % shorter computation times than Wolf Pack Algorithm (WPA) and Particle Swarm Optimization (PSO), respectively. Compared with conventional algorithms, the framework achieves faster convergence and better economic performance, demonstrating strong applicability for two-stage sequential scheduling in future NTPS environments.