The inherent variability and intermittency of photovoltaic (PV) power generation pose significant challenges to accurate forecasting, which is crucial for the secure and economic operation of power systems with high renewable penetration. Accurate PV power forecasting remains challenging because existing decomposition-based hybrids usually lack component-specific modeling and adaptive fusion, while many deep models either underexploit feature-temporal interactions or incur high computational overhead. To address these issues, this paper proposes WDFNet, a Wavelet-Based Dual-Branch Network with Adaptive Fusion for multivariate PV power forecasting. WDFNet first uses discrete wavelet decomposition to decouple the input series into trend and seasonal components. A lightweight predictor is then used to model the trend component, while the seasonal component is processed by a dedicated branch enhanced with a Feature-Temporal Module and a Gated Correction Module. Finally, the outputs of the two branches are adaptively fused through a cross-attention mechanism. Experiments on six PV datasets with forecasting horizons of 24, 48, 96, and 192 steps demonstrate the superior overall performance of WDFNet. Specifically, on Site 1, WDFNet achieves the best performance, outperforming the second-best model by 9.7% in terms of MSE, underscoring its effectiveness in PV power forecasting.

Multivariate time series forecasting plays a key role in a variety of scenarios, including weather, electricity, economics, and other domains. Models based on Multilayer Perceptrons (MLPs) offer advantages such as simplicity of structure and low computational complexity. However, many MLP-based methods tend to overlook the inter-variable (or cross-channel) dependencies, which are essential for accurate predictions. Understanding these inter-variable relationships is crucial for improving prediction performance. To address this challenge, we introduce the Kernel improved Spectral Theory Model (KiST): a model that leverages kernel mapping to enhance the separation between samples to improve model generalization. Additionally, KiST incorporates spectral theory by designing a learnable intermediate matrix to capture and fuse the relationships between channels and sequences. Our empirical studies show that the proposed KiST model achieves state-of-the-art performance in multivariate time series forecasting tasks. Notably, KiST exhibits advantages in parameter efficiency and inference speed. We also present comprehensive ablation experiments and visualizations to validate the model’s effectiveness.

Accurate multivariate time series (MTS) forecasting is crucial for applications in traffic planning, energy management, financial investment, and healthcare. The challenge of forecasting MTS lies in managing the complex temporal dynamics and inter-channel relationships. However, despite the progress achieved by previous studies, they still fall short of adeptly addressing these complexities, leaving substantial scope for further refinement. To fill this gap, this paper proposes DiM, which seamlessly integrates the difference-inverted (DI) embedding strategy and the multi-head graph learning mechanism within the Metaformer framework. Specifically, the DI embedding employs a straightforward differencing operation to compensate for the limitations of previous inverted embedding methods in capturing temporal dynamics, thereby enhancing the model's ability to discern temporal patterns without significantly increasing computational complexity. The multi-head graph learning mechanism dynamically adjusts multiple graph structures to better represent the evolving relationships between channels, surpassing the constraints of static graph structures typically used in GNNs. The effectiveness of the DiM has been validated across eleven public datasets, where it achieved the best results in 53 out of 55 mean absolute error metrics compared to existing state-of-the-art models, establishing a new benchmark in MTS forecasting. Code is available at https://github.com/Yipengmo/DiM.

This study presents a hydraulic high-performance engineered cementitious composite (HHP-ECC) incorporating LC3 composites with advages of high strength, great tensile ductility, excellent freeze-thaw and water flow abrasion resistence. A systematic investigation was conducted to evaluate the influence of LC3 composite incorporation on the performance of high-performance engineered cementitious composites (HHP-ECC). The study focused on key parameters including workability, mechanical properties, microstructure evolution, and autogenous shrinkage, with substitution ratios of LC3 ranging from 0 % to 45 %. Additionally, the freeze-thaw resistance and abrasion resistance of HHP-ECC were analyzed. The findings reveal that the inclusion of LC3 composites could mitigate the plastic viscosity of fresh paste. Both the ultimate tensile strength and compressive strength first increased and then decreased, whereas the ultimate tensile strain showed a continuous upward trend with the increasing incorporation of LC3 composites. A minimal autogenous shrinkage could be achieved when 30 % of cement was replaced by LC3 composites. Moreover, all HHP-ECC specimen has a freeze-thaw grade of F500, attributed to the presence of significant gel pores within the matrix. The abrasion process of HHP-ECC was captured by the 3D scanning technology. Compared to hydraulic C40 and UHPC, the LC3-30 sample demonstrated significantly higher abrasion resistance strength and lower abrasion depth since its large tensile strain energy that could efficiently absorb the abrasion energy exerted on the abrasion hollows.

Abrasion damage caused by water flow with solid particles is a critical durability problem for hydraulic concrete structures. However, conventional evaluation methods, such as total mass loss or average wear depth, cannot capture the spatially heterogeneous nature of surface abrasion within a single specimen. This study investigates the intra-specimen relationship between local surface morphology and local abrasion behavior using cementitious composites without coarse aggregates. A local abrasion resistance strength metric was proposed to quantify abrasion resistance at the surface-unit scale, and a multi-scale morphology framework was established to characterize local geometric features over different spatial ranges. The results show that local abrasion resistance is affected by the synergistic effects of multi-scale surface morphology rather than by any single roughness parameter. In particular, morphology features at an intermediate scale comparable to the abrasive medium size showed the strongest contribution, indicating a scale-matching effect between surface undulation and abrasive action. Partial Least Squares (PLS)-Ridge models further demonstrated the predictive advantage of the proposed descriptor system, increasing the average coefficient of determination (R2) from 0.258 for conventional descriptor groups to 0.880. The proposed framework provides a spatially resolved basis for identifying abrasion-prone local weak zones on hydraulic concrete surfaces and supports the evaluation and design of abrasion-resistant cement-based materials under severe hydraulic conditions.

Electricity load is highly volatile and uncertain, which makes accurate forecasting very challenging but critical for effective power management and grid stability. Existing forecasting methods struggle to handle longterm time series, complex multi-scale temporal patterns, and noisy data, which significantly limits their effectiveness in real-world applications. To address these challenges, this paper proposes a novel Multi-Resolution Graph Neural Network (MRGNN) that integrates multi-scale patching and Frequency-enhanced Direct Forecast (FreDF) approach. MRGNN partitions input sequences into multi-resolution segments to capture both short-term fluctuations and long-term trends while leveraging graph neural networks to model spatial-temporal dependencies. Additionally, a frequency-enhanced loss function aligns predictions with ground truth in both time and frequency domains, mitigating label correlation issues. Experiments on real-world datasets demonstrate MRGNN's superiority among all models, achieving top performance in 92 out of 96 test cases, particularly for long-term forecasting horizons. The results highlight MRGNN's robustness and precision, offering practical value for optimizing power systems and advancing spatio-temporal forecasting research.

In the domain of multivariate time series analysis, the concept of channel independence has been increasingly adopted, demonstrating excellent performance due to its ability to eliminate noise and the influence of irrelevant variables. However, such a concept often simplifies the complex interactions among channels, potentially leading to information loss. To address this challenge, we propose a strategy of channel independence followed by mixing. Based on this strategy, we introduce CSformer, a novel framework featuring a two-stage multiheaded self-attention mechanism. This mechanism is designed to extract and integrate both channel-specific and sequence-specific information. Distinctively, CSformer employs parameter sharing to enhance the cooperative effects between these two types of information. Moreover, our framework effectively incorporates sequence and channel adapters, significantly improving the model's ability to identify important information across various dimensions. Extensive experiments on several real-world datasets demonstrate that CSformer achieves state-of-the-art results in terms of overall performance.

Debris flows and landslides are geological hazards that pose a significant threat to life and property. To reduce the impact of these disasters and improve management efficiency, there is an urgent need to establish flexible and effective post-disaster management methods. High-precision aerial images offer significant advantages in this field, as their mobility and high resolution provide more detailed terrain information compared to traditional satellite images, especially in post-disaster emergency response scenarios. This paper proposes a systematic solution to detect and classify geohazard areas in high-precision aerial images. The approach uses the Attention-Pyramid U-Net (APU-Net) for aerial image segmentation, enabling accurate localization and identification of geohazard areas. For APU-Net, in the encoder part, the Atrous Spatial Pyramid Pooling (ASPP) is added in parallel to the encoder module of each layer and then connected in series with the Squeeze and Excitation (SE) module. In the decoder part, the Convolutional Block Attention Module (CBAM) is introduced after each upsampling. The APU-Net model is used to segment aerial images and extract key features. These features are then used to classify debris flow and landslide events. The experimental results show that the proposed method achieves an accuracy of 85.44% for the detection of geohazard areas and an accuracy of 76.13% for the classification of debris flows and landslides under complex terrain conditions. These results demonstrate that the proposed method provides robust technical support for disaster management by effectively detecting and classifying debris flows and landslides in high-precision aerial images.

In the context of the artificial intelligence revolution, the demand for long-term time series forecasting (LTSF) across various applications continues to rise. Contemporary deep learning models such as Transformer-based and MLP-based models have shown promise. However, these state-of-the-art (SOTA) approaches encounter notable limitations: Transformer-based models suffer from low computational efficiency and the inherent restrictions of point-wise attention mechanisms, while MLP-based models struggle to effectively capture local temporal dependencies. To overcome these challenges, this paper introduces a novel lightweight architecture centered around CNN-based models with an inherent receptive field, GLCN, explicitly designed to capture and discern intricate relationships in time series. The architecture features a key component, the global-local block, which initially segments the time series into subseries levels to preserve the underlying semantic information of temporal variations and subsequently captures both inter- and intra-patch inherent global and local temporal dynamics. In particular, GLCN utilizes a lightweight CNN-based architecture for prediction to significantly enhance training speed by 65.1% and 86.0% on the Weather and ETTh1 datasets, respectively, while reducing parameters by 94.8% and 94.4%. Comprehensive experiments on seven real-world datasets demonstrate that GLCN reduces contemporary SOTA approaches by 1.6% and 1.8% in Mean Squared Error and Mean Absolute Error.

Cavitation erosion poses a significant challenge in the durability design of spillway lining concrete. With the advent of low-carbon, high-performance cementitious composites, traditional concrete is increasingly being replaced by high-performance fiber-reinforced cementitious composites (HPFRCC). However, the cavitation erosion performance of these novel materials remains largely unexplored. This study systematically investigates the cavitation erosion behavior of HPFRCC and high-strength mortar through ultrasonic cavitation erosion tests. HPFRCC is reinforced with steel fibers and comprises a quaternary cementitious matrix that includes three types of solid waste materials along with titanium slag sand. The findings demonstrate that HPFRCC exhibits lower volume/mass loss, reduced loss rates, smaller fractal dimensions, and a slower rate of erosion depth development compared to mortar. Furthermore, the presence of steel fibers and the enhanced strength of the matrix result in a distinct cavitation erosion evolution process in HPFRCC relative to mortar. The incorporation of steel fibers effectively slows the rate of erosion depth development, while the fiber-matrix interface performance and fiber orientation significantly influence the efficacy of the steel fibers. Additionally, a positive correlation exists between the compressive strength of both HPFRCC and mortar and their resistance to cavitation erosion. However, when the compressive strength exceeds 140 MPa, the effect of compressive strength on the cavitation erosion resistance of cement-based composites diminishes significantly. The findings reveal the superior cavitation erosion resistance of HPFRCC, providing practical insights for the further application of novel cementitious materials.

This research systematically investigates the compression responses and damage evolution of high-performance engineered cementitious composites (HP-ECC) under triaxial monotonic and cyclic compressive loads. A thorough analysis was conducted on the influence of lateral confining stress, ranging from 1 to 20 MPa, on the mechanical characteristics and performance degeneration of HP-ECC. The findings reveal that the application of lateral confining stress significantly enhances the mechanical properties of HP-ECC, particularly in terms of peak stress and peak strain, while exerting minimal influence on the elastic modulus of the matrix. Moreover, the failure surface of HP-ECC, encompassing the compression and tensile meridians along with the deviatoric plane, was effectively characterized using a five-parameter failure criterion. Under cyclic loading, the accumulation of plastic strain exhibits a linear correlation with the envelope unloading strain, decreasing as lateral confining stress increases. Although significant stiffness degradation occurs with an increasing number of loading cycles, this deterioration is mitigated under higher levels of confining stress. Consequently, a robust exponential function incorporating confining stress and unloading strain parameters was proposed to model the damage evolution of HP-ECC. Additionally, an elastoplastic damage model was introduced to accurately characterize the triaxial monotonic and cyclic stress-strain behavior, demonstrating high precision in predicting experimental results from both this study and the existing literature.

Multivariate time series (MTS) forecasting is extensively applied in real-world scenarios. Recent studies introduced the concept of channel independence, which has achieved significant results. However, this approach often misses the intricate interrelations among features and lacks physical interpretation. To address this problem, we propose a new framework, the Feature-Temporal block, designed to extract both temporal and feature information. Each block contains two modules: a feature module using a gating mechanism to understand competitive interactions among features, and a temporal module using learnable filters for frequency domain filtering. This innovative design allows FTMLP to combine information across feature and sequence dimensions, as well as time and frequency domains. The computational efficiency of FTMLP is enhanced due to its exclusive use of Multi-Layer Perceptron (MLP). Comprehensive experiments on several real-world datasets demonstrate that FTMLP achieves state-of-the-art results in terms of overall performance. Code is available at https://github.com/whxlearning/FTMLP.

The aligned CNT-based composite in transverse direction to the alignment is considered as completely insulating and unsuitable for utilization in de-icing system. Here, we have found that the aligned CNT-based composite in transverse direction to the alignment exhibited excellent electrothermal performance at high voltage, for the first time. Furthermore, we compared and investigated the electrical and electrothermal performance of the aligned and random non-percolative CNT/crosslinked polyethylene (XLPE) composites at high voltage. A temperature increase of 7.5-72.7 celcius at a record high voltage of 900-3100 V is reached for the aligned CNT/XLPE composites. The filed emission at high voltage enables the electron to transfer along the CNTs radial direction, resulting in joule heating for the aligned CNT/XLPE composites in transverse direction to the alignment. Besides, the non-percolative structure contributes a temperature increase of 8.2-40.1 celcius at a applied voltage of 900-2800 V to random CNT/XLPE composites. The high applied voltages is significantly higher than rGO-based composites. More importantly, effective de-icing and anti-icing performance at high voltage of 900-2800 V are achieved for the random non-percolative and aligned CNT-based composites at-20 celcius. These results demonstrate the po-tential of applying the CNT/XLPE composites in de/anti-icing system of transmission-lines and other fields requiring joule heating at high voltage.

This study proposes a serviceable hydraulic high-performance-engineered cementitious composites (HHP-ECC) incorporating LC3 composites based on the performance requirements of the submerged bridge pier concrete subjected to freeze-thaw damage and water flow abrasion in Qinghai-Tibet plateau. Firstly, the impact of LC3 composites on the hydration kinetics, workability, mechanical characteristics, microstructure, and autogenous shrinkage of HHP-ECC blending varying substitution ratio ranging from 0 to 45%, has been investigated systematically. Besides, both freeze-thaw resistance and abrasion resistance of HHP-ECC were specially researched. The results indicate that the incorporating LC3 composites delayed the hydration process, while mitigated the plastic viscosity of fresh paste. The ultimate tensile strength and compressive strength slightly increased and then reduced, while the ultimate tensile strain increased as LC3 composites increased. A minimal autogenous shrinkage could be achieved when 30 % of cement in the Ref group was replaced by LC3 composites. Moreover, all HHP-ECC specimen has a freeze-thaw grade of F500 since amounts of gel pores pore were presented in the matrix. Compared to hydraulic C40 and traditional UHPC, LC3-30 sample exhibited much higher abrasion resistance strength and far lower abrasion depth because the abrasion hollows on LC3-30 specimen can mostly withstand the abrasion energy behaving on the concrete surface due to its high strength and high toughness. LC3-30 sample is selected as a promising matrix for the bridge pier in the Qinghai-Tibet plateau due to its high compressive and flexural strength, high ductility, low autogenous shrinkage, great freeze-thaw resistance, and extraordinary abrasion resistance, which also provides an environmentally friendly solution for cement industries.

Time series is a special type of sequence data, a sequence of real-valued random variables collected at even intervals of time. The real-world multivariate time series comes with noises and contains complicated local and global temporal dynamics, making it difficult to forecast the future time series given the historical observations. This work proposes a simple and effective framework, coined as TimeSQL, which leverages multi-scale patching and smooth quadratic loss (SQL) to tackle the above challenges. The multi-scale patching transforms the time series into two-dimensional patches with different length scales, facilitating the perception of both locality and long-term correlations in time series. SQL is derived from the rational quadratic kernel and can dynamically adjust the gradients to avoid overfitting to the noises and outliers. Theoretical analysis demonstrates that, under mild conditions, the effect of the noises on the model with SQL is always smaller than that with MSE. Based on the two modules, TimeSQL achieves new state-of-the-art performance on the eight real-world benchmark datasets. Further ablation studies indicate that the key modules in TimeSQL could also enhance the results of other models for multivariate time series forecasting, standing as plug-and-play techniques.

In geological scene registration with laser-scanned point cloud data, traditional algorithms often face reduced precision and efficiency due to extensive data volume and scope, which increase complexity and computational demands. This study introduces, to our knowledge, a novel registration method to address these limitations. Through dimension reduction that integrates height and curvature data, this approach converts point clouds into images, streamlining feature extraction. Log-variance enhancement mitigates information loss from dimensionality reduction, aiding in coarse registration. Further, incorporating weighted distances of feature points into the Iterative Closest Point (ICP) algorithm improves precision in point matching. Experiments indicate an average threefold increase in initial registration efficiency compared to traditional coarse registration algorithms, with improvements in accuracy. The optimized ICP algorithm achieves 50% and 15% accuracy improvements across various datasets, enhancing large-scale geological point cloud data registration. (c) 2024 Optica Publishing Group

AbstractThe pre‐insertion resistors (PIR) within high‐voltage circuit breakers are critical components and warm up by generating Joule heat when an electric current flows through them. Elevated temperature can lead to temporary closure failure and, in severe cases, the rupture of PIR. To accurately predict the temperature of PIR, this study combines finite element simulation techniques with Support Vector Regression (SVR) optimized by an Improved Whale Optimization Algorithm (IWOA) approach. The IWOA includes Tent mapping, a convergence factor based on the sigmoid function, and the Ornstein–Uhlenbeck variation strategy. The IWOA‐SVR model is compared with the SSA‐SVR and WOA‐SVR. The results reveal that the prediction accuracies of the IWOA‐SVR model were 90.2% and 81.5% (above 100°C) in the ± 3°C temperature deviation range and 96.3% and 93.4% (above 100°C) in the ± 4°C temperature deviation range, surpassing the performance of the comparative models. This research demonstrates that the method proposed can realize the online monitoring of the temperature of the PIR, which can effectively prevent thermal faults PIR and provide a basis for the opening and closing of the circuit breaker within a short period.

Wind Power Forecasting has emerged as a critical and dynamic research area in response to the growing demand for renewable energy. The unpredictable and stochastic nature of wind conditions, encompassing factors such as wind speed, wind direction, air temperature, and barometric pressure, poses unique challenges for accurate forecasting of wind power generation. Reliable wind power generation forecasts are essential for optimizing energy grid management, ensuring grid stability, and facilitating the integration of wind energy with existing power systems. To address these challenges, this research introduces Powerformer, a Transformer-based model designed to improve the accuracy of wind power prediction. Powerformer utilizes the infrastructure of the Transformer with innovative modifications to address the complexity of wind power prediction, enhancing temporal feature extraction capabilities while reducing complexity. The research in this study includes a comprehensive set of experiments, revealing that Powerformer achieves superior results among all models. Furthermore, the model exhibits stronger robustness, as confirmed through a series of ablation experiments validating the reasonableness of the model design.

The inherent volatility and intermittency of wind power present significant forecasting challenges, undermining the efficient integration of wind energy into the power grid. Existing methodologies, notably long short-term memory (LSTM) networks, encounter significant limitations due to their inefficiencies in processing long sequences, difficulties in capturing multi-scale temporal dynamics, and heightened sensitivity to noisy data, which can severely hamper model performance. To address these challenges, This paper proposes the frequency filter enhanced dual LSTM network (FDNet), a novel approach that directly addresses the constraints of the LSTM and improves the accuracy and stability of wind power forecasting. Specifically, FDNet employs the patching operation to divide the original time series into several sub-sequences, potentially boosting the computational efficiency. Furthermore, a specific frequency filter is designed and incorporated into FDNet, effectively reducing the influence of noise. Finally, a dual LSTM structure is employed, which enables FDNet to adeptly discover both short-term local temporal patterns and long-term global temporal patterns inherent in wind power data. Extensive experiments across three datasets demonstrate that FDNet significantly outperforms existing methods, achieving up to 11.0% reduction in mean absolute error and 8.1% in root mean squared error on the HL dataset, underscoring its effectiveness in wind power forecasting.

To enhance the stability and speed of surface crack detection, a detection algorithm combining PeleeNet and YOLOv3 is proposed. PeleeNet framework is used to replace Darknet-53 framework of YOLOv3 so as to effectively integrate different local features and improve the detection rate. The feature attention module is integrated into the PeleeNet’s framework to highlight the saliency of the crack detection in the image, and through the receptive field module RFB broaden the effective field of view of the network, and increase the detection accuracy of small targets. Instead of the standard convolution, the depth separable convolution is employed to reduce the amount of parameter calculation in the feature pyramid network. Then, the CIoU loss function is introduced to strengthen the classification and regression accuracy of the model. The experimental results on the fracture data set show that AP 50 and AP 75 reach 97.68% and 77.87% respectively, 8.4% and 12.4% higher than the original YOLOv3. Meanwhile, the detection speed reaches 30 frames per second with the size of model parameters being only 30% of YOLOv3. As can be seen, the PeleeNet＿yolov3 lightweight model proposed has produced an obvious effect on the detection of crack targets, with a small amount of calculations and parameters involved. Suitable for mobile terminal system, the study presents a great value of application especially for small volume, low power consumption and low computing power computing platforms.