Enhancing temperature prediction in the UAE: a process-driven framework for adaptive learning with GRU-CNN hybrid models

Modeling Earth Systems and Environment, Journal Year: 2025, Volume and Issue: 11(2)

Published: Jan. 22, 2025

Language: Английский

---

Time Series Forecasting Model Based on the Adapted Transformer Neural Network and FFT-Based Features Extraction

Sensors, Journal Year: 2025, Volume and Issue: 25(3), P. 652 - 652

Published: Jan. 23, 2025

In today’s data-driven world, where information is one of the most valuable resources, forecasting behavior time series, collected by modern sensor networks and IoT systems, crucial across various fields, including finance, climatology, engineering. However, existing neural network models often struggle with series different sensors due to challenges such as large data volumes, long-term dependencies, noise, anomalies, which can negatively impact predictive accuracy. This paper aims enhance accuracy proposing an adapted transformer architecture combined innovative preprocessing method. The proposed technique employs fast Fourier transform (FFT) transition from domain frequency domain, enriching additional frequency-domain features. These features are represented complex numbers, improve informational content for subsequent analysis, thereby boosting performance. Furthermore, introduces a modified model specifically designed address identified in prediction. performance was evaluated using three diverse datasets sensors, each varying measurement frequencies, types, application domains, providing comprehensive comparison state-of-the-art LSTM, FFT-LSTM, DeepAR, Transformer, FFT-Transformer. Extensive evaluation five distinct metrics demonstrates that consistently outperforms methods, achieving highest all datasets.

Language: Английский

---

Hybrid Transformer Model with Liquid Neural Networks and Learnable Encodings for Buildings’ Energy Forecasting

Energy and AI, Journal Year: 2025, Volume and Issue: unknown, P. 100489 - 100489

Published: Feb. 1, 2025

Language: Английский

---

An integrated approach for prediction of magnitude using deep learning techniques

Neural Computing and Applications, Journal Year: 2024, Volume and Issue: 36(27), P. 16991 - 17006

Published: June 4, 2024

Language: Английский

---

Leveraging Hybrid Deep Learning Models for Enhanced Multivariate Time Series Forecasting

Neural Processing Letters, Journal Year: 2024, Volume and Issue: 56(5)

Published: Aug. 23, 2024

Time series forecasting is crucial in various domains, ranging from finance and economics to weather prediction supply chain management. Traditional statistical methods machine learning models have been widely used for this task. However, they often face limitations capturing complex temporal dependencies handling multivariate time data. In recent years, deep emerged as a promising solution overcoming these limitations. This paper investigates how learning, specifically hybrid models, can enhance address the shortcomings of traditional approaches. dual capability handles intricate variable interdependencies non-stationarities forecasting. Our results show that achieved lower error rates higher $$R^2$$ values, signifying their superior predictive performance generalization capabilities. These architectures effectively extract spatial features dynamics by combining convolutional recurrent modules. study evaluates architectures, On two real-world datasets - Traffic Volume Air Quality TCN-BiLSTM model best overall performance. For Volume, an score 0.976, Quality, it reached 0.94. highlight model's effectiveness leveraging strengths Temporal Convolutional Networks (TCNs) multi-scale patterns Bidirectional Long Short-Term Memory (BiLSTMs) retaining contextual information, thereby enhancing accuracy

Language: Английский

---

Global and Local Approaches for Forecasting of Long-Term Natural Gas Consumption in Poland Based on Hierarchical Short Time Series

Energies, Journal Year: 2024, Volume and Issue: 17(2), P. 347 - 347

Published: Jan. 10, 2024

This study presents a novel approach for predicting hierarchical short time series. In this article, our objective was to formulate long-term forecasts household natural gas consumption by considering the structure of territorial units within country’s administrative divisions. For purpose, we utilized data from Poland. The length series an important determinant set. We contrast global techniques, which employ uniform method across all series, with local methods that fit distinct each Furthermore, compare conventional statistical machine learning (ML) approach. Based on analyses, devised forecasting exhibit exceptional performance. have demonstrated models provide better than models. Among ML models, neural networks yielded best results, MLP network achieving comparable performance LSTM while requiring significantly less computational time.

Language: Английский

---

Optimizing differentiated podcasts to promote students’ self-regulation and engagement, self-efficacy and performance in asynchronous learning

Journal of Research in Innovative Teaching & Learning, Journal Year: 2024, Volume and Issue: 17(2), P. 368 - 390

Published: June 29, 2024

Purpose Our study focuses on providing empirical evidence regarding the optimization of podcasting in asynchronous learning. This action research aimed to innovate delivery classes using differentiated podcasts. Design/methodology/approach We utilized as design for study. Participating entails developing practical knowledge improve educational practices through specific methods and critical perspectives (Sáez Bondía Cortés Gracia, 2022). According Burns (2007), involves deliberate interventions usually prompted by identified issues, mysteries or inquiries that individuals social setting seek change. Implementing changes enhance individuals' actions understanding within their context is focus (Kemmis, 2010). The study’s approach ideal examining new gaining enhanced theoretical insights (Altrichter et al ., 2002). Engaging helps empowers us impact continuous reflection, exploration action. Through this iterative process, we can continuously our comprehension make substantial strides toward fostering positive transformation. Findings findings showed an apparent rise student regulation engagement remarkable enhancements learning outcomes, demonstrated differences pre-test final exam scores. These results highlight actual effect specialized podcasts self-paced inducing students' self-efficacy provides valuable effectively incorporating into education, offering innovations improvement practice among educators institutions adapting ever-changing landscape environment while catering diverse needs learners. pioneering various styles environments. Research limitations/implications Although current sample offered insights, upcoming studies could gain from more extensive participant groups strengthen reliability guarantee broader applicability across demographics contexts. Moreover, length intervention may have been relatively brief, which limited ability evaluate long-term customized results. Continued investigation effects these provide effectiveness over time help shape creation lasting approaches. Practical implications Innovation teaching attuned students. Social promotes equitable eventually outcomes Originality/value created tailored

Language: Английский

---

Dynamic Real-Time Prediction of Reclaimed Water Volumes Using the Improved Transformer Model and Decomposition Integration Technology

Sustainability, Journal Year: 2024, Volume and Issue: 16(15), P. 6598 - 6598

Published: Aug. 1, 2024

In recent years, wastewater reuse has become crucial for addressing global freshwater scarcity and promoting sustainable water resource development. Accurate inflow volume predictions are essential enhancing operational efficiency in treatment facilities effective utilization. Traditional decomposition integration models often struggle with non-stationary time series, particularly peak anomaly sensitivity. To address this challenge, a differential model based on real-time rolling forecasts been developed. This uses an initial prediction machine learning (ML) model, followed by using Complete Ensemble Empirical Mode Decomposition Adaptive Noise (CEEMDAN). A Time-Aware Outlier-Sensitive Transformer (TS-Transformer) is then applied integrated predictions. The ML-CEEMDAN-TSTF demonstrated superior accuracy compared to basic ML models, other Transformer-based models. hybrid explicitly incorporates time-scale differentiated information as feature, improving the model’s adaptability complex environmental data predictive performance. TS-Transformer was designed make more sensitive anomalies peaks issues such anomalous data, uncertainty suboptimal forecasting accuracy. results indicated that: (1) introduction of significantly enhanced accuracy; (2) higher ML-CEEMDAN-Transformer; (3) TS-Transformer-based consistently outperformed those LSTM eXtreme Gradient Boosting (XGBoost). Consequently, research provides precise robust method predicting reclaimed volumes, which holds significant implications clean environment management.

Language: Английский

---

Advanced Deep Learning Techniques for Battery Thermal Management in New Energy Vehicles

Energies, Journal Year: 2024, Volume and Issue: 17(16), P. 4132 - 4132

Published: Aug. 19, 2024

In the current era of energy conservation and emission reduction, development electric other new vehicles is booming. With their various attributes, lithium batteries have become ideal power source for vehicles. However, lithium-ion are highly sensitive to temperature changes. Excessive temperatures, either high or low, can lead abnormal operation batteries, posing a threat safety entire vehicle. Therefore, developing reliable efficient Battery Thermal Management System (BTMS) that monitor battery status prevent thermal runaway becoming increasingly important. recent years, deep learning has gradually widely applied in fields as an method, it also been some extent BTMS. this work, we discuss basic principles related optimization elaborate on algorithmic principles, frameworks, applications advanced methods We several emerging algorithms proposed feasibility BTMS applications. Finally, obstacles faced by potential directions development, proposing ideas progress. This paper aims analyze technologies commonly used provide insights into combination technology trams assist

Language: Английский

---

Evaluating the Effectiveness of Time Series Transformers for Demand Forecasting in Retail

Mathematics, Journal Year: 2024, Volume and Issue: 12(17), P. 2728 - 2728

Published: Aug. 31, 2024

This study investigates the effectiveness of Transformer-based models for retail demand forecasting. We evaluated vanilla Transformer, Informer, Autoformer, PatchTST, and temporal fusion Transformer (TFT) against traditional baselines like AutoARIMA AutoETS. Model performance was assessed using mean absolute scaled error (MASE) weighted quantile loss (WQL). The M5 competition dataset, comprising 30,490 time series from 10 stores, served as evaluation benchmark. results demonstrate that significantly outperform baselines, with TFT leading metrics. These achieved MASE improvements 26% to 29% WQL reductions up 34% compared seasonal Naïve method, particularly excelling in short-term forecasts. While Autoformer PatchTST also surpassed methods, their slightly lower, indicating potential further tuning. Additionally, this highlights a trade-off between model complexity computational efficiency, models, though computationally intensive, offering superior forecasting accuracy slower AutoARIMA. findings underscore approaches enhancing forecasting, provided demands are managed effectively.

Language: Английский

---