Fluctuating Environmental Conditions (FECs) are a critical barrier to accurate photovoltaic (PV) power forecasting. Existing models often fail to capture abrupt and stochastic fluctuations, leading to reduced forecasting reliability. To address this challenge, this study proposes an interpretable Photovoltaic Knowledge-Informed Neural Network (PKINN). The framework incorporates a Quadratic Explicit Model (QEM) to derive explicit expressions of PV power and transparently capture abrupt variations, while a Fluctuation Allocation Mechanism (FAM) employs a fluctuation sensitivity coefficient to quantify fluctuation intensity and allocate input data to specialized prediction branches. The proposed PKINN framework enables adaptive learning across diverse FECs and enhances forecasting performance. Experimental evaluations on two types of PV modules demonstrate that PKINN reduces the root mean square error by at least 8.73% compared with state-of-the-art models across diverse FECs.

Electricity prices are high-level indicators of modern energy system operations. Accurate modeling of market dynamics and electricity price forecasting (EPF) are therefore essential for the stability and efficiency of power markets. However, traditional models struggle to dynamically capture multiple-scenario influencing factors, leading to degraded predictive precision under high-volatility market conditions. To address this issue, this study proposes a Reasoning-Informed Semantic-Large Language Model (RIS-LLM) framework. It introduces a Reasoning Feature Extraction (RFE) module that dynamically identifies volatility-sensitive features through adaptive decomposition. By constructing a causal-volatility matrix, the RFE enables the model to reason across multiple driving factors, distinguishing transient fluctuations from structural trends. A Time Series Semantics (TSS) alignment mechanism is designed to semantically describe Granger-causal dependencies between environmental factors and electricity price dynamics. It employs a predictive-bias attention strategy to align volatility-driven features with interpretable semantic labels, such as demand surge or renewable oversupply, thereby highlighting the key determinants of electricity price fluctuations. Based on these, the RIS-LLM framework establishes a reasoning pathway from volatility features to semantic correlation structures. It enhances the model's ability to reason over multiple driving factors, enabling accurate forecasting of electricity price under dynamic market conditions. Experiments on a public benchmark dataset demonstrate that RIS-LLM achieves state-of-the-art forecasting accuracy. It delivers improvements in predictive accuracy while exhibiting stable performance even under highly volatile conditions.

Due to the variability of environmental parameters, Photovoltaic (PV) systems frequently operate under Noisy Operating Conditions (NOC). The existing Flexible Power Point Tracking (FPPT) algorithms encounter significant challenge in misjudging the optimal voltage points under NOC. This paper presents a Frequency Domain-Deep Q-Network (FD-DQN) approach to FPPT by incorporating frequency domain analysis through the Fourier transform, which decomposes irradiance signals to capture dynamic changes caused by noise. Additionally, a novel Signal Quality Factor (SQF) is introduced to quantify noise and refine the learning process, minimizing overfitting in the presence of noisy data. By analyzing the influence of different frequency components, the FD-DQN enables more accurate tracking with fewer misjudgments. Experimental results indicate that the proposed approach achieves a minimum improvement of 5% in tracking accuracy and reduces misjudgments by around 50% under NOC, outperforming traditional methods in both simulated and experimental scenarios with fluctuating irradiance.

Object detection via visible-infrared imagery fusion is an important technology in automatic driving, video surveillance and field fire monitoring. We present SWF-DETR, a novel multi-modal image fusion and detection framework using infrared and visble image. The proposed framework leverages the advanced Transformer architecture in both fusion and detection stages. This design aligns the feature Spaces of the two stages and alleviates the differences usually caused by the use of different network architectures. In the fusion stage, the full color fusion map of YCbCr is used to retain richer features, thereby reducing the fluctuation of detection accuracy in various types of objects. In addition, the algorithm utilizes a validated loss function to retain the necessary information throughout the fusion procedure and to maintain the fused features’ conformance to the detector’s representation specifications, ensuring optimal convergence. Comprehensive studies on two publicly available benchmark datasets demonstrate that the SWF-DETR algorithm not only improves mAP50 by $1.4 \%$ over existing algorithms but also exhibits greater stability, evidenced by a standard deviation of 11.2 across categories. This marks a $9.7 \%$ stability improvement compared to the previous best of 12.4.

The solar tracking system is one of the effective methods to enhance Photovoltaic (PV) power generation efficiency. However, existing systems face challenges in managing power losses when PV panels experience partial shading, resulting in prolonged tracking times and reduced average power output. In this study, we propose a sensorless Beta-Particle-Filter (BPF) solar tracking method that introduces a Beta parameter to define a restricted search area, thereby avoiding unnecessary global exploration. Additionally, a shadow identification process is incorporated, allowing the system to dynamically adjust the initial tracking range according to the shading level, thereby significantly reducing search time. Simulations and experiments demonstrate that the proposed solar tracking method increases the power generation by 60% under the Partial Shading Condition (PSC) compared to the fixed PV panel and achieves an 8% improvement in power generation compared to the latest particle filter method.

Moving-object perception must decide which image regions correspond to real motion and keep every instance identified over time. Methods that read motion from appearance, optical flow, or estimated trajectories lose that evidence under poor illumination, adverse weather, reflections, and occlusion. Radar is a natural remedy because it measures radial velocity directly instead of inferring it from photometric correspondence. However, existing benchmarks do not jointly provide radar measurements, dense moving-instance masks, and temporally consistent identities for surveillance. We therefore introduce RGBTR-Motion, a synchronized and calibrated fixed-camera benchmark that pairs RGB, thermal, and radar streams with dense instance masks and temporally consistent identities across diverse surveillance scenes. We also develop SAM-Radar, an RGB, thermal, and radar-based segmentation and tracking framework built on SAM 3. SAM-Radar's radar-aware detector fuses calibrated RGBT features with radar returns that are grounded at their projected image locations, and motion supervision, implemented as foreground classification of those projected returns, teaches the detector to reject clutter without any text prompt. The tracker associates accepted radar returns with individual trajectories and uses them as physical evidence that a visually degraded target remains present. This allows it to bridge short periods of low visibility or occlusion and reconnect a reappearing target to its existing identity instead of starting a new track. SAM-Radar attains 0.7027 IoU and 0.8090 F1-50, and raises MOTA, HOTA, and IDF1 by 0.2977, 0.1603, and 0.2857 over the strongest competing values.

The Earth's revolution and geographic variability introduce spatial uncertainty in photovoltaic (PV) systems. Subtle spatial variations give rise to dynamic shading conditions (DSC), which disrupt power prediction over time. Existing models often neglect to capture the effects of spatial uncertainty, and consequently struggle to address the DSC in PV systems. This paper presents a 3D-PV framework, which introduces a deblurring 3D reconstruction technique to produce spatial representations, preserving details of PV panels and their surrounding environment. Further, shadow variation matrices are constructed by the proposed ComputeShader-based shadow calculation algorithm, serving as a spatio-temporal representation to bridge the obtained spatial representations and dynamic shading variations. Building on the spatio-temporal representations, 3D-PV performs semantic fusion of shadow dynamics and irradiance signals, enabling temporally consistent power prediction under DSC. Experimental results, including ablation studies, demonstrate that precise spatial modeling effectively captures and simulates accurate shadow patterns over time. In particular, 3D-PV outperforms state-of-the-art prediction methods, achieving a 23.95 % reduction in mean squared error (MSE) for prediction accuracy. These results highlight the benefits of explicitly modeling spatial uncertainty and dynamically fusing spatio-temporal representations with irradiance signals under DSC, enabling accurate prediction of PV power.

Different from resistive loads, constant power loads (CPLs) may threaten the stability of power interfaces (PIs) due to the negative incremental impedance. As one of the promising PIs, dual active bridge (DAB) converters have received widespread attention. However, the large-signal stability criteria of DAB converters under CPLs remain challenging. To fill this gap, this article thoroughly explores the inherent stability mechanism by analyzing the state trajectory of the DAB converter for the first time. Sufficient conditions for the large-signal stability of the closed-loop controlled DAB converter under CPLs are specified. Based on the derived trajectory operation criteria of DAB converters, an advanced boundary control in the geometrical domain is proposed to guarantee the large-signal stability of DAB converters feeding CPLs. The main design considerations and recommendations are given. Experimental results show that the proposed strategy can reduce the start-up transient time by 50% and speed up the output-voltage-reference-change transient time by over 30% compared to the traditional control while ensuring a stable operation of DAB converters under CPLs.

With the rapid advancements of sensor technology and deep learning, autonomous driving systems are providing safe and efficient access to intelligent vehicles as well as intelligent transportation. Among these equipped sensors, the radar sensor plays a crucial role in providing robust perception information in diverse environmental conditions. This review focuses on exploring different radar data representations utilized in autonomous driving systems. Firstly, we introduce the capabilities and limitations of the radar sensor by examining the working principles of radar perception and signal processing of radar measurements. Then, we delve into the generation process of five radar representations, including the ADC signal, radar tensor, point cloud, grid map, and micro-Doppler signature. For each radar representation, we examine the related datasets, methods, advantages and limitations. Furthermore, we discuss the challenges faced in these data representations and propose potential research directions. Above all, this comprehensive review offers an in-depth insight into how these representations enhance autonomous system capabilities, providing guidance for radar perception researchers. To facilitate retrieval and comparison of different data representations, datasets and methods, we provide an interactive website at https://radar-camera-fusion.github.io/radar.

Underwater depth estimation is crucial for marine applications such as autonomous navigation and robotics. However, monocular depth estimation in underwater environments remains challenging due to the rapid attenuation of the red light spectrum in deep waters, causing bluish-green color distortion, while suspended particles and limited illumination lead to blurry effects. These underwater degradations severely affect the performance of RGB-based depth estimation methods, particularly in background regions. To overcome the limitations of color-based depth estimation techniques in underwater scenarios, this paper proposes a novel dual-source depth fusion framework leveraging color and light attenuation information. First, an innovative input space is designed inspired by the principle of depth-dependent light transmission in underwater environments. This input space enhances robustness against color distortion and improves the capacity to capture depth information, particularly in blurry underwater regions. Subsequently, we develop an adaptive fusion module to optimize the strengths of both RGB and this new input space across varying underwater conditions. This module employs a novel confidence-based mechanism to dynamically assess the reliability of depth information from each source on a per-pixel basis. By leveraging a learned confidence map, it can adaptively weigh and fuse the contributions of RGB and the new input space. This strategy enables optimal depth estimation across diverse underwater scenarios. Extensive experiments on multiple challenging datasets demonstrate that our method consistently outperforms current state-of-the-art monocular depth estimation techniques in various subaqueous environments.

Accurate photovoltaic (PV) power prediction is crucial for effective energy management, particularly in regions with variable weather conditions. This paper presents the Holographic PV Physical Model-LSTM Frequency Network (HPPM-LFNet) model, which integrates a Holographic PV Physical Model (HPPM) framework for physical power output prediction through detailed panel-level shading analysis with an LSTM Frequency Network (LFNet) for uncertainty modeling. The HPPM component employs 3D environmental modeling to generate physically-certified power output under partial shading conditions, while the LFNet incorporates frequency-domain analysis to enhance adaptability to rapid weather changes. Experimental evaluation demonstrates that HPPM-LFNet consistently outperforms traditional methods, achieving superior performance across different time scales forecasting with $R^{2}$ values exceeding 0.97 under diverse weather conditions.

Blockchain has started moving to Ethereum, decentralised and open-source. One of the most important application domains of the Ethereum blockchain is decentralised finance (DeFi). It facilitates speed-up trading, in absence of the traditional financial intermediaries, such as brokerages, exchanges, or banks. The three main types of digital asset trading strategies are concentrated liquidity, unbounded liquidity, and grid trading. Concentrated liquidity has recently been designed to increase the liquidity provision that indicates the convertibility of assets. Currently, the performance comparison of these three types of trading remains largely unclear. This research proposes back-testing algorithms to measure their return on investment (ROI). Our research has been conducted on real Ethereum blockchains and has discovered that their ROIs vary and depend on price fluctuation. These findings will shed new insights in the design of future decentralised trading strategies.

Specularity poses significant challenges in computer vision (CV), often leading to performance degradation in various tasks. Despite its importance, the CV field lacks a comprehensive review of specularity detection techniques. This survey addresses this gap by synthesizing diverse definitions of specularity and providing a unified framework to enhance consistency. It also presents a systematic review of traditional and deep learning-based methods for detecting specularity. Comparative experiments on a standardized dataset enable in-depth evaluation of each method, highlighting their strengths and limitations. The survey further provides structured insights and guidance for selecting appropriate methods across diverse scenarios. Through this, it identifies key areas for future research, aiming to support the development of more advanced detection models. By integrating diverse methodologies and quantitative analyzes, this survey contributes to a deeper understanding of current advancements and potential innovations in specularity detection.

Predicting the performance of photovoltaic (PV) systems is crucial for optimizing renewable energy utilization. However, traditional time-series methods focus only on temporal patterns, overlooking environmental variations, while dynamic conditions such as partial shading further complicate power prediction. To address this shading-induced variability, we propose a Temporal and Environment-Informed Prediction (TEIP) framework, which enhances PV power prediction by dynamically structuring temporal and environmental data through a novel multi-spatial attention LSTM (MSAL) network. This framework utilizes the TE matrix to capture structured environmental conditions over time, including the variability caused by partial shading. A dual-branch MSAL model uniquely processes environmental data through spatial feature extraction, which is then sequentially processed by LSTM to capture temporal dependencies. This hierarchical spatial-temporal processing enables dynamic adaptation to changing environmental conditions. Experimental results show the framework achieves superior prediction accuracy with R2 of 0.952 under sunny conditions, significantly outperforming traditional approaches. The framework demonstrates exceptional robustness by maintaining consistent performance (R2 of 0.948) even under challenging cloudy conditions, validating its effectiveness for real-world applications.

Dynamic scene reconstruction is essential in robotic minimally invasive surgery, providing crucial spatial information that enhances surgical precision and outcomes. However, existing methods struggle to address the complex, temporally dynamic nature of endoscopic scenes. This paper presents ST-Endo4DGS, a novel framework that models the spatio-temporal volume of dynamic endoscopic scenes using unbiased 4D Gaussian Splatting (4DGS) primitives, parameterized by anisotropic ellipses with flexible 4D rotations. This approach enables precise representation of deformable tissue dynamics, capturing intricate spatial and temporal correlations in real time. Additionally, we extend spherindrical harmonics to represent time-evolving appearance, achieving realistic adaptations to lighting and view changes. A new endoscopic normal alignment constraint (ENAC) further enhances geometric fidelity by aligning rendered normals with depth-derived geometry. Extensive evaluations show that ST-Endo4DGS outperforms existing methods in both visual quality and real-time performance, establishing a new state-of-the-art in dynamic scene reconstruction for endoscopic surgery.

In the Virtual Reality (VR) gaming industry, maintaining immersion during real-world interruptions remains a challenge, particularly during transitions along the reality-virtuality continuum (RVC). Existing methods tend to rely on digital replicas or simple visual transitions, neglecting to address the aesthetic discontinuities between real and virtual environments, especially in highly stylized VR games. This paper introduces the Environment-Aware Stylized Transition (EAST) framework, which employs a novel style-transferred 3D Gaussian Splatting (3DGS) technique to transfer real-world interruptions into the virtual environment with seamless aesthetic consistency. Rather than merely transforming the real world into game-like visuals, EAST minimizes the disruptive impact of interruptions by integrating real-world elements within the framework. Qualitative user studies demonstrate significant enhancements in cognitive comfort and emotional continuity during transitions, while quantitative experiments highlight EAST's ability to maintain visual coherence across diverse VR styles.