Search Results (130)

Accurately predicting future activities in egocentric (first-person) videos is a challenging yet essential task, requiring robust object recognition and reliable forecasting of action patterns. However, the limited number of observable frames in such videos often lacks critical semantic context, making long-term predictions particularly difficult. Traditional approaches, especially those based on recurrent neural networks, tend to suffer from cumulative error propagation over extended time steps, leading to degraded performance. To address these challenges, this paper introduces a novel framework, Virtual Frame-Augmented Guided Forecasting (VFGF), designed specifically for long-term egocentric activity prediction. The VFGF framework enhances semantic continuity by generating and incorporating virtual frames into the observable sequence. These synthetic frames fill the temporal and contextual gaps caused by rapid changes in activity or environmental conditions. In addition, we propose a Feature Guidance Module that integrates anticipated activity-relevant features into the recursive prediction process, guiding the model toward more accurate and contextually coherent inferences. Extensive experiments on the EPIC-Kitchens dataset demonstrate that VFGF, with its interpolation-based temporal smoothing and feature-guided strategies, significantly improves long-term activity prediction accuracy. Specifically, VFGF achieves a state-of-the-art Top-5 accuracy of 44.11% at a 0.25 s prediction horizon. Moreover, it maintains competitive performance across a range of long-term forecasting intervals, highlighting its robustness and establishing a strong foundation for future research in egocentric activity prediction. Full article

Squall lines are some of the most common types of mesoscale cloud systems in tropical and subtropical regions. Thunderstorms associated with these systems are among the major causes of weather-related disasters and socio-economic losses in many regions across the world. This study investigates the capability of the Weather Research and Forecasting (WRF) model in simulating squall line features over the South African Highveld region. Two squall line cases were selected based on the availability of South African Weather Service (SAWS) weather radar data: 21 October 2017 (early austral summer) and 31 January–1 February 2018 (late austral summer). The European Centre for Medium-Range Weather Forecasts ERA5 datasets were used as observational proxies to analyze squall line features and compare them with WRF simulations. Mid-tropospheric perturbations were observed along westerly waves in both cases. These perturbations were coupled with surface troughs over central interior together with the high-pressure systems to the south and southeast of the country creating strong pressure gradients over the plateau, which also transports relative humidity onshore and extending to the Highveld region. The 2018 case also had a zonal structured ridging High, which was responsible for driving moisture from the southwest Indian Ocean towards the eastern parts of South Africa. Both ERA5 and WRF captured onshore near surface (800 hPa) winds and high-moisture contents over the eastern parts of the Highveld. A well-defined dryline was observed and well simulated for the 2017 event, while both ERA5 and WRF did not show any dryline for the 2018 case that was triggered by orography. While WRF successfully reproduced the synoptic-scale processes of these extreme weather events, the simulated rainfall over the area of interest exhibited a broader spatial distribution, with large-scale precipitation overestimated and convective rainfall underestimated. Our study shows that models are able to capture these systems but with some shortcomings, highlighting the need for further improvement in forecasts. Full article

This study evaluates and corrects ECMWF precipitation forecasts (Set VI-ENS extended) over the confluence of Asian monsoons and westerlies, deriving a time series of correction factors for medium- and long-term hydrological forecasting. Based on a 15-year dataset (2008–2023), a dominant spatial and temporal bias pattern was identified: ~50% of the study area—in particular, the entire Tibetan Plateau—experienced overestimated precipitation, with larger relative errors in dry seasons than in wet seasons. Daily correction factors were derived using the linear scaling method and applied to distributed hydrological models for the Mekong, Salween, and Brahmaputra river basins. The results demonstrated substantial efficacy in correcting streamflow forecasts, particularly in the Brahmaputra basin at the Nuxia station, where the relative error in the total water volume over a 32-day period was reduced from 25% to 10% during the calibration period (2008–2020) and from 20% to 9% in the validation period (2021–2023). Furthermore, over 90% (calibration) and 85% (validation) of hydrological forecast events were successfully corrected at Nuxia. Comparable improvements were observed in key stations across the Salween and Mekong basins, with the combined success rates exceeding 70% and 65%, demonstrating the method’s regional robustness. Challenges remain in areas with weak linear relationships between forecasted and observed data, highlighting the need for further investigation. Full article

Electric buses are key in the strategy towards a greenhouse-gas-neutral fleet. However, their restrictions in terms of range and refueling as well as their increased price point present new challenges for public transport companies. This study aims to address, based on real-world operational data, how energy consumption and charging behavior affect battery aging and how operational strategies can be optimized to extend battery life under realistic conditions. This article presents an energy consumption analysis with respect to ambient temperatures and average vehicle speed based exclusively on real-world data of an urban bus fleet, providing a data foundation for range forecasting and infrastructure planning optimized for public transport needs. Additionally, the State of Charge (SOC) window during operation and vehicle idle time as well as the charging power were analyzed in this case study to formulate recommendations towards a more battery-friendly treatment. The central research question is whether battery-friendly operational strategies—such as reduced charging power and lower SOC windows—can realistically be implemented in daily public transport operations. The impact of the recommendations on battery lifetime is estimated using a battery aging model on drive cycles. Finally, the reduction in CO₂ emissions compared to diesel buses is estimated. Full article

This paper presents an advanced time-series forecasting methodology that integrates multiple machine learning models to improve data prediction in enhanced long-range navigation (eLoran) systems. The analysis evaluates five forecasting approaches: multivariate linear regression, long short-term memory (LSTM) networks, random forest (RF), a fusion model combining LSTM and RF, and a dynamic weighting (DW) model. The results demonstrate that the DW model achieves the highest prediction accuracy while maintaining strong computational efficiency, making it particularly suitable for real-time applications with stringent performance requirements. Although the LSTM model effectively captures temporal dependencies, it demands considerable computational resources. The hybrid model utilises the strengths of LSTM and RF to enhance the accuracy but involves extended training times. By contrast, the DW model dynamically adjusts the relative contributions of LSTM and RF on the basis of the data characteristics, thereby enhancing the accuracy while significantly reducing the computational demands. Demonstrating exceptional performance on the ASF2 dataset, the DW model provides a well-balanced solution that combines precision with operational efficiency. This research offers valuable insights into optimising additional secondary phase factor (ASF) prediction in eLoran systems and highlights the broader applicability of real-time forecasting models. Full article

The National Institute of Earth Physics (NIEP) in Romania has upgraded its seismic monitoring stations into multifunctional platforms equipped with advanced devices for measuring gas emissions, magnetic fields, telluric fields, solar radiation, and more. This enhancement enabled the integration of a seismic forecasting system designed to extend the alert time of the existing warning system, which previously relied solely on seismic data. The implementation of an Operational Earthquake Forecast (OEF) aims to expand NIEP’s existing Rapid Earthquake Early Warning System (REWS) which currently provides a warning time of 25–30 s before an earthquake originating in the Vrancea region reaches Bucharest. The AFROS project (PCE119/4.01.2021) introduced fundamental research essential to the development of the OEF system. As a result, real-time analyses of radon and CO₂ emissions are now publicly available at afros.infp.ro, dategeofizice. The primary monitored area is Vrancea, known for producing the most destructive earthquakes in Romania, with impacts extending to neighboring countries such as Bulgaria, Ukraine, and Moldova. The structure and methodology of the monitoring network are adaptable to other seismic regions, depending on their specific characteristics. All collected data are stored in an open-access database available in real time, geobs.infp.ro. The monitoring methods include threshold-based event detection and seismic data analysis. Each method involves specific technical nuances that distinguish this monitoring network as a novel approach in the field. In conclusion, experimental results indicate that the Gutenberg-Richter law, combined with gas emission measurements (radon and CO₂), can be used for real-time earthquake forecasting. This approach provides warning times ranging from several hours to a few days, with results made publicly accessible. Another key finding from several years of real-time monitoring is that the value of fundamental research lies in its practical application through cost-effective and easily implementable solutions—including equipment, maintenance, monitoring, and data analysis software. Full article

This study investigates the theoretical foundations and practical advantages of fractional-order optimization in computational machine learning, with a particular focus on stock price forecasting using long short-term memory (LSTM) networks. We extend several widely used optimization algorithms—including Adam, RMSprop, SGD, Adadelta, FTRL, Adamax, and Adagrad—by incorporating fractional derivatives into their update rules. This novel approach leverages the memory-retentive properties of fractional calculus to improve convergence behavior and model efficiency. Our experimental analysis evaluates the performance of fractional-order optimizers on LSTM networks tasked with forecasting stock prices for major companies such as AAPL, MSFT, GOOGL, AMZN, META, NVDA, JPM, V, and UNH. Considering four metrics (Sharpe ratio, directional accuracy, cumulative return, and MSE), the results show that fractional orders can significantly enhance prediction accuracy for moderately volatile stocks, especially among lower-cap assets. However, for highly volatile stocks, performance tends to degrade with higher fractional orders, leading to erratic and inconsistent forecasts. In addition, fractional optimizers with short-memory truncation offer a favorable trade-off between computational efficiency and modeling accuracy in medium-frequency financial applications. Their enhanced capacity to capture long-range dependencies and robust performance in noisy environments further justify their adoption in such contexts. These results suggest that fractional-order optimization holds significant promise for improving financial forecasting models—provided that the fractional parameters are carefully tuned to balance memory effects with system stability. Full article

The increasing complexity of power grids and integration of renewable energy sources necessitate accurate power load forecasting across multiple time horizons. While existing methods have advanced significantly, they often struggle with consistent performance across different prediction ranges, leading to suboptimal resource allocation. We propose MP-RWKV (Multi-Path Recurrent Weighted Key–Value), an enhanced architecture that builds upon RWKV-TS and addresses these challenges through parallel processing paths for temporal modeling. Our model maintains robust performance across both short-term and long-term forecasting scenarios through its context state mechanism and position-aware attention. Evaluated on extensive power load data, MP-RWKV demonstrates superior performance over state-of-the-art baselines, including Transformer-based models and LSTM variants. The model achieves the lowest Mean Absolute Error (MAE) across prediction horizons ranging from 24 h to 432 h, showing particular strength in maintaining consistent accuracy where traditional models deteriorate. Notably, MP-RWKV successfully balances immediate temporal correlations with extended dependencies, offering promising implications for power grid management and sustainable energy systems. The model’s stable performance across varying prediction horizons makes it particularly suitable for real-world power load forecasting applications. Full article

Accurate prediction of Cloud Liquid Water Content (CLWC) is critical for understanding and forecasting weather phenomena, particularly in regions with complex microclimates. This study integrates high-resolution ERA5 climatic data from the European Centre for Medium-Range Weather Forecasts (ECMWF) with radiosonde observations from the Hong Kong area to address data accuracy and resolution challenges. Machine learning (ML) models—specifically Fine Tree regressors—were employed to interpolate radiosonde data, resolving temporal and spatial discrepancies and enhancing data coverage. A metaheuristic algorithm was also applied for data cleansing, significantly improving correlations between input features (temperature, pressure, and humidity) and CLWC. The methodology was tested across multiple ML algorithms, with ensemble models such as Bagged Trees demonstrating superior predictive accuracy and robustness. The approach substantially improved CLWC profile reliability, outperforming traditional methods and addressing the nonlinear complexities of atmospheric data. Designed for scalability, this methodology extends beyond Hong Kong’s unique conditions, offering a flexible framework for improving weather prediction models globally. By advancing CLWC estimation techniques, this work contributes to enhanced weather forecasting and atmospheric science in diverse climatic regions. Full article

Ocean forecasting is critical for various applications and is essential for understanding air–sea interactions, which contribute to mitigating the impacts of extreme events. While data-driven forecasting models have demonstrated considerable potential and speed, they often primarily focus on spatial variations while neglecting temporal dynamics. This paper presents the TSformer, a novel non-autoregressive spatio-temporal transformer designed for medium-range ocean eddy-resolving forecasting, enabling forecasts of up to 30 days in advance. We introduce an innovative hierarchical U-Net encoder–decoder architecture based on 3D Swin Transformer blocks, which extends the scope of local attention computation from spatial to spatio-temporal contexts to reduce accumulation errors. The TSformer is trained on 28 years of homogeneous, high-dimensional 3D ocean reanalysis datasets, supplemented by three 2D remote sensing datasets for surface forcing. Based on the near-real-time operational forecast results from 2023, comparative performance assessments against in situ profiles and satellite observation data indicate that the TSformer exhibits forecast performance comparable to leading numerical ocean forecasting models while being orders of magnitude faster. Unlike autoregressive models, the TSformer maintains 3D consistency in physical motion, ensuring long-term coherence and stability. Furthermore, the TSformer model, which incorporates surface auxiliary observational data, effectively simulates the vertical cooling and mixing effects induced by Super Typhoon Saola. Full article

Sea surface temperature (SST) is crucial for weather forecasting, climate modeling, and environmental monitoring. This study proposes a novel prediction model that achieves a 60-day forecast with a root mean square error (RMSE) consistently below 0.9 °C. The model combines the nonlinear feature extraction of a deep belief network (DBN) with the high-precision regression of support vector regression (SVR), enhanced by spatiotemporal secondary calibration (SSC) to better capture SST variation patterns. Using satellite-derived remote sensing data, the DBN-SVR model outperforms baseline methods in both the Indian Ocean and North Pacific regions, demonstrating strong applicability across diverse marine environments. This work advances long-term SST prediction capabilities, providing a reliable foundation for extended-range marine forecasts. Full article

Recent studies have identified a near six-year oscillation (SYO) in Global Navigation Satellite Systems (GNSS) surface displacements, with a degree 2, order 2 spherical harmonic (SH) pattern and retrograde motion. The cause is uncertain, with proposals ranging from deep Earth to near-surface sources. This study investigates the SYO and possible causes from surface loading. Considering the irregular spatiotemporal distribution of GNSS data and the variety of contributors to surface displacements, we used synthetic experiments to identify optimal techniques for estimating low degree SH patterns. We confirm a reported retrograde SH degree 2, order 2 displacement using GNSS data from the same 35 stations used in a previous study for the 1995–2015 period. We also note that its amplitude diminished when the time span of observations was extended to 2023, and the retrograde dominance became less significant using a larger 271-station set. Surface loading estimates showed that terrestrial water storage (TWS) loads contributed much more to the GNSS degree 2, order 2 SYO, than atmospheric and oceanic loads, but TWS load estimates were highly variable. Four TWS sources—European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5), Modern-Era Retrospective analysis for Research and Applications (MERRA), Global Land Data Assimilation System (GLDAS), and Gravity Recovery and Climate Experiment (GRACE/GRACE Follow-On)—yielded a wide range (24% to 93%) of predicted TWS contributions with GRACE/GRACE Follow-On being the largest. This suggests that TWS may be largely responsible for SYO variations in GNSS observations. Variations in SYO GNSS amplitudes in the extended period (1995–2023) were also consistent with near surface sources. Full article

This paper critically examines artificial intelligence applications in stock market forecasting, addressing significant gaps in the existing literature that often overlook the tension between theoretical market efficiency and empirical predictability. While numerous reviews catalog methodologies, they frequently fail to rigorously evaluate model performance across different market regimes or reconcile statistical significance with economic relevance. We analyze techniques ranging from traditional statistical models to advanced deep learning architectures, finding that ensemble methods like Extra Trees, Random Forest, and XGBoost consistently outperform single classifiers, achieving directional accuracy of up to 86% in specific market conditions. Our analysis reveals that hybrid approaches integrating multiple data sources demonstrate superior performance by capturing complementary market signals, yet many models showing statistical significance fail to generate economic value after accounting for transaction costs and market impact. By addressing methodological challenges including backtest overfitting, regime changes, and implementation constraints, we provide a novel comprehensive framework for rigorous model assessment that bridges the divide between academic research and practical implementation. This review makes three key contributions: (1) a reconciliation of the Efficient Market Hypothesis with AI-driven predictability through an adaptive market framework, (2) a multi-dimensional evaluation methodology that extends beyond classification accuracy to financial performance, and (3) an identification of promising research directions in explainable AI, transfer learning, causal modeling, and privacy-preserving techniques that address current limitations. Full article

Offshore wind farms play a pivotal role in the global transition to clean energy but remain susceptible to diverse meteorological hazards—ranging from highly variable wind speeds and temperature anomalies to severe oceanic disturbances—that can jeopardize both turbine safety and overall power output. Although Hidden Markov Models (HMMs) have a longstanding track record in operational forecasting, this study leverages and extends their capabilities by introducing a dynamic HMM framework tailored specifically for multi-risk offshore wind applications. Building upon historical datasets and expert assessments, the proposed model begins with initial transition and observation probabilities and then refines them adaptively through periodic or event-triggered recalibrations (e.g., Baum–Welch), thus capturing evolving weather patterns in near-real-time. Compared to static Markov chains, naive Bayes classifiers, and RNN (LSTM) baselines, our approach demonstrates notable accuracy gains, with improvements of up to 10% in severe weather conditions across three industrial-scale wind farms. Additionally, the model’s minutes-level computational overhead for parameter updates and state decoding proves feasible for real-time deployment, thereby supporting proactive scheduling and maintenance decisions. While this work focuses on the core dynamic HMM method, future expansions may incorporate hierarchical structures, Bayesian uncertainty quantification, and GAN-based synthetic data to further enhance robustness under high-dimensional measurements and rare, long-tail meteorological events. In sum, the multi-risk forecasting methodology presented here—though built on an established HMM concept—offers a practical, adaptive solution that significantly bolsters safety margins and operational reliability in offshore wind power systems. Full article

This paper presents an innovative approach to wind energy forecasting through the implementation of an extended long short-term memory (xLSTM) model. This research addresses fundamental limitations in time-sequence forecasting for wind energy by introducing architectural enhancements to traditional LSTM networks. The xLSTM model incorporates two key innovations: exponential gating with memory mixing and a novel matrix memory structure. These improvements are realized through two variants, i.e., scalar LSTM and matrix LSTM, which are integrated into residual blocks to form comprehensive architectures. The xLSTM model was validated using SCADA data from wind turbines, with rigorous preprocessing to remove anomalous measurements. Performance evaluation across different wind speed regimes demonstrated robust predictive capabilities, with the xLSTM model achieving an overall coefficient of determination value of 0.923 and a mean absolute percentage error of 8.47%. Seasonal analysis revealed consistent prediction accuracy across varied meteorological patterns. The xLSTM model maintains linear computational complexity with respect to sequence length while offering enhanced capabilities in memory retention, state tracking, and long-range dependency modeling. These results demonstrate the potential of xLSTM for improving wind power forecasting accuracy, which is crucial for optimizing turbine operations and grid integration of renewable energy resources. Full article