With the rise of time-of-use and tiered electricity pricing, energy consumers are encouraged to adopt peak-shifting strategies by automatically controlling high-power appliances. These help lower energy costs while enhancing the power grid's stability. To support such energy management with high resilience and responsiveness, reliable short-term load forecasting (STLF) plays a critical role. STLF predicts electricity consumption over time horizons ranging from minutes to days, using historical data, temporal patterns, and contextual factors. Traditional top-down forecasting methods struggle to capture the complex consumption patterns of diverse and mixed appliance loads. Although bottom-up methods improve forecasting accuracy by integrating appliance-level data, monitoring all appliances is costly, and many do not meaningfully impact total load prediction. Therefore, we propose GCA-BULF, a bottom-up short-term load forecasting framework based on grouped critical appliances, supported by three key designs. First, the Critical Appliance Filtering module ranks appliances according to their power consumption, switching frequency, and usage pattern periodicity, and identifies critical ones through iterative load decomposition. Next, the Related Appliance Grouping module clusters these appliances based on spatial and temporal correlations for group-level forecasting. Finally, the Collaborative Load Forecasting module refines the total load prediction by combining multiple group-level forecasts. We evaluate GCA-BULF on residential and office building load forecasting tasks. Experimental results reveal that GCA-BULF improves hourly total load forecasting by 20.85%-57.88% compared to existing top-down methods and by 33.03%-92.48% compared to bottom-up methods.

Recent breakthroughs in Multimodal Large Language Models (MLLMs) have gained significant recognition within the deep learning community, where the fusion of the Video Foundation Models (VFMs) and Large Language Models(LLMs) has proven instrumental in constructing robust video understanding systems, effectively surmounting constraints associated with predefined visual tasks. These sophisticated MLLMs exhibit remarkable proficiency in comprehending videos, swiftly attaining unprecedented performance levels across diverse benchmarks. However, their operation demands substantial memory and computational resources, underscoring the continued importance of traditional models in video comprehension tasks. In this paper, we introduce a novel learning paradigm termed MLLM4WTAL. This paradigm harnesses the potential of MLLM to offer temporal action key semantics and complete semantic priors for conventional Weakly-supervised Temporal Action Localization (WTAL) methods. MLLM4WTAL facilitates the enhancement of WTAL by leveraging MLLM guidance. It achieves this by integrating two distinct modules: Key Semantic Matching (KSM) and Complete Semantic Reconstruction (CSR). These modules work in tandem to effectively address prevalent issues like incomplete and over-complete outcomes common in WTAL methods. Rigorous experiments are conducted to validate the efficacy of our proposed approach in augmenting the performance of various heterogeneous WTAL models.

The high penetration of renewable energy presents significant challenges for the power grid in balancing supply and demand. Demand response (DR) is crucial for stable grid operation, yet existing research often overlooks actual electricity consumption patterns for the following day and the diversity among user types, compromising assessment accuracy and applicability. To address these shortcomings, this paper proposes an integrated load regulation system comprising “Identification-Prediction-Scheduling”. First, we identify specific user groups and utilize load prediction models to accurately forecast future loads. By analyzing predicted loads alongside historical data, we uncover electricity consumption patterns and DR potential for the following day, providing better insights for power grid scheduling. Finally, we implement a load control strategy aimed at optimizing grid stability. Our results show that the load forecasting model achieves average errors of 17.86% and 16.16% at intervals of 15 minutes and 1 hour, respectively. When demand is set at 1500 kW, the proposed load control strategy predicts a response load curtailment of 2216.85 kW and a DPI score of 50.75, significantly outperforming both the random selection strategy (1887.11 kW, DPI score of 18.39) and the priority to large FBC strategy (288.86 kW, DPI score of 40.78). This approach ensures better enterprise selection, contractual compliance, and improved benefits for electrical users.

Using extensive training data from SA-1B, the Segment Anything Model (SAM) has demonstrated exceptional generalization and zero-shot capabilities, attracting widespread attention in areas such as medical image segmentation and remote sensing image segmentation. However, its performance in the field of image manipulation detection remains largely unexplored and unconfirmed. There are two main challenges in applying SAM to image manipulation detection: a) reliance on manual prompts, and b) the difficulty of single-view information in supporting cross-dataset generalization. To address these challenges, we develops a cross-view prompt learning paradigm called IMDPrompter based on SAM. Benefiting from the design of automated prompts, IMDPrompter no longer relies on manual guidance, enabling automated detection and localization. Additionally, we propose components such as Cross-view Feature Perception, Optimal Prompt Selection, and Cross-View Prompt Consistency, which facilitate cross-view perceptual learning and guide SAM to generate accurate masks. Extensive experimental results from five datasets (CASIA, Columbia, Coverage, IMD2020, and NIST16) validate the effectiveness of our proposed method.

Weakly supervised temporal action localization aims to learn to locate actions in videos from video-level or point-level labels, avoiding the need for costly frame-level annotations. Unlike previous work that relies solely on visual modality information, we propose incorporating audio information into the weakly supervised temporal action localization task. While audio-visual localization tasks combine audio and visual information for video localization, temporal action localization often deals with action categories that have weak audio cues. To address this, we propose EAV-Mamba, the first audio-visual perception modeling method based on Mamba. Leveraging Mamba’s powerful audio-visual perception capabilities, we developed modules such as Audio-Perceptive Flow Enhancement, Audio-Perceptive RGB Enhancement, and Audio Self-Perceptive Enhancement. Extensive experiments on two publicly available temporal action localization datasets demonstrate that EAV-Mamba achieves efficient audio-visual perception modeling and state-of-the-art performance in weakly supervised temporal action localization tasks.

High-precision full-waveform inversion is a key method for subsurface parameter modeling. Enhancing the contribution of weak low-frequency signals to low-wavenumber components is crucial for inversion. Existing frequency normalization schemes can significantly leverage low-frequency contributions in inversion, but their discrete normalization approach produces pronounced high-wavenumber noise. We propose a deep-learning-parameterized frequency normalization scheme that effectively enhancing low-frequency contribution and suppressing high-wavenumber artifacts. By innovatively constructing a deep-learning-driven inversion framework with frequency-normalized adjoint gradients: it uses a deep learning network to parameterize model parameters, solves the time-domain solver with a normalized source, builds a gradient without frequency crosstalk, and forms a cyclic network parameter update process. Through a multi-scale learning optimization scheme, the final inversion result is output via the deep network. Synthetic data results demonstrate that the proposed method enhances low-wavenumber modeling capability and suppresses high-wavenumber noise; its application to land data confirms its ability to bridge the medium-wavenumber gap between the initial background model and migration imaging.

The application of deep learning techniques to full waveform inversion (FWI) theory represents a significant research direction. Leveraging the nonlinear representations offered by deep learning and conducting practical FWI are paramount. This article utilizes the classic adjoint method in FWI to compute the gradients of model parameters, employing deep learning to represent model parameters and optimize network training. The focus is on achieving high-precision FWI through multiscale deep learning optimization. Specifically, deep neural networks are used to represent model parameter information and compute gradients of model parameters on high-performance platforms. The gradients of the network parameters are automatically obtained through backpropagation, with deep learning optimization tools updating the network parameters and, consequently, the model parameters. To enhance inversion accuracy, a multiscale learning strategy is introduced, where deep networks optimize the learning of model parameter information at each scale, ensuring effective representation of inversion parameter information across multiple scales. Experimental results demonstrate that deep learning reparameterization methods possess broad-spectrum modeling capabilities. The multiscale deep learning strategy significantly improves inversion accuracy, and the reparameterization method of deep learning shows potential for high-precision modeling under conditions of sparse and noisy observational data. Furthermore, the application of field data underscores the reliability of the proposed method.

The natural damping effect of the Earth typically results in significant distortion of seismic waveforms, which greatly diminishes the precise of subsequent processes such as parameter inversion, migration imaging, and reservoir description. Compensating for this attenuation is crucial to achieving precise underground parameter measurements. While inversion or imaging techniques that rely on wave path compensation have the potential to address attenuation effects better, they encounter challenges, including heightened demands for input models, rapidly escalating algorithm intricacy, and computational burdens. Consequently, developing novel attenuation compensation methods that balance computational efficiency and accuracy is important in enhancing the precision of exploring complex reservoirs. This study utilizes a groundbreaking convolutional neural network (CNN), which integrates an attention mechanism and residual learning. This network establishes an inherent link between attenuated seismic data and their nonattenuated counterparts, effectively accomplishing data-driven compensation for seismic data attenuation. The more advanced acoustic (nonattenuated) full-waveform inversion (FWI) or reverse time migration framework is directly applied to enhance the modeling or imaging of attenuated seismic data with improved accuracy and efficiency. Simulation data and actual test results confirm that the suggested Q-compensation approach successfully enhances the amplitude of deep structural reflection signals, rectifies phase distortion induced by attenuation, and widens the seismic frequency range. This mitigates issues such as the cycle-skipping problem associated with low-frequency absence in traditional FWI and the numerical instability and increased computational complexity found in attenuation compensation FWI. Furthermore, the imaging profile’s resolution is further heightened due to the effective attenuation correction and enhancement of high-frequency components.

Time-domain elastic full waveform inversion (FWI) uses seismic data to recover the high-resolution subsurface medium properties for structural imaging and lithologic identification. The approximate pulse generated by the explosion source in seismic exploration evolves into a limited bandwidth seismic wavelet through the propagation of the seismic wave in the underground medium, exhibiting strong energy near the dominant frequency and weak energy far away. Therefore, time-domain FWI using the band-limited seismic wavelet can only match the energy of the dominant frequencies present in a dataset to a large extent, resulting in insufficient low-wavenumber updates of model parameters because of the weak energy of low frequencies. To remove the effect of finite-frequency-band wavelet spectra from the time-domain FWI, we propose a time-domain elastic FWI without wavelet spectral limitation. In our method, a newly refined seismic wavelet with a normalized amplitude and accurate phase is used to propagate seismic waves in the time domain. Data residual measurement is performed based on the summation of single-frequency residuals between the normalized synthetic and observed frequencies. The adjoint-state method is used to approximate the gradients of the model parameters, and the decoupled wavefields obtained by the phase-sensitive detection method were involved in the gradient calculation. Overall, the proposed method helps FWI to avoid falling into local minima by enhancing the low-wavenumber reconstruction of the velocity models. The elastic FWI numerical tests and field data FWI application demonstrate that our approach can reliably recover high-precision inversion results.

Full-waveform inversion (FWI) serves as a useful tool to quantitatively investigate the properties of the subsurface. Presently, 3-D elastic FWI uses a finite-difference time-domain (FDTD) approach in numerical simulation. However, such an FDTD scheme often includes only second-order temporal approximations, causing errors in temporal dispersion in the case of a large time-stepping size. Such temporal dispersion will affect the inversion results and reduce the inversion quality. We introduce a unique 3-D elastic FWI using a temporal fourth-order finite-difference (FD) approximation. A new quasi-stress–velocity elastic equation is solved by the temporal fourth-order and spatial arbitrary even-order FDTD method, and a novel inversion procedure for the convolutional objective function based on this equation is derived. The multiscale strategy is used to enhance the robustness of our algorithm. The forward modeling and FWI examples presented here demonstrate that our method can achieve modeling and inversion with a high degree of accuracy.

Seismic deconvolution is a typical ill-posed inverse problem. The regularization technique in terms of different prior information is used for a unique and stable solution. Due to the difference between prior information and the actual subsurface situation, it is hard to obtain a solution with satisfactory accuracy and resolution. This letter presents the dictionary learning as an efficient adaptive deconvolution method for the reflectivity reconstruction problem. Considering the curse of dimensionality of conventional dictionary learning and the suboptimal solution of the patch-based dictionary learning, we take the convolutional sparse coding (CSC) model as the dictionary learning method. In this method, the prior information can be obtained from the well-log data in the form of sparse CSC dictionary of reflectivity. On the assumption that the deposition of the subsurface layers is stable, the CSC dictionary extracted from the well-log data can also be applied in the whole work area. The CSC-based deconvolution can be seen as the adaptive deconvolution due to the independence of the assumption made about the reflectivity and seismic data. The process of the adaptive CSC-based deconvolution is divided into three parts. First, the CSC dictionary is learned from the well-log data. Then, the objective function is formulated by combining the CSC dictionary and the single-channel seismic record misfit term for the reconstruction of reflectivity. Finally, the objective function is efficiently solved with the coordinate descent approach. We illustrate the performance of our adaptive deconvolution with synthetic and field seismic data.

Elastic full waveform inversion (FWI) is more suitable to process multicomponent seismic data and can provide more subsurface medium information than acoustic FWI often with lower efficiency. Except for the parallel algorithms, source-encoding methods are usually adopted to improve the efficiency of FWI, but it often includes crosstalk noise. Besides, the additional source estimation process, critical for a successful FWI, would counteract the high-efficiency advantage of the source-encoding algorithm. We propose an elastic FWI with source-independent crosstalk-free encoding algorithm to solve the above problems. Arbitrary-phase harmonic sine functions are used as new source wavelets to perform the time-domain wavefield simulation regardless of the true wavelet. Treating the harmonic wavelet as the encoding operator and based on the orthogonality of trigonometric functions within integer periods, the amplitude and phase of each source are recovered from the blended source and adjoint wavefields so that the influence of crosstalk noise is avoided. With the deblended data, the proposed algorithm can be naturally applied to unfixed-spread acquisition systems. Moreover, we can conveniently perform the multiscale inversion by controlling the frequencies of simultaneous-source signals as conventional frequency-domain FWI does. Synthetic examples show that the proposed algorithm has high efficiency and accuracy with a strong robustness to the incorrect wavelets.