Innovative machine learning approaches for indoor air temperature forecasting in smart infrastructure

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Янв. 2, 2025

Abstract Efficient energy management and maintaining an optimal indoor climate in buildings are critical tasks today’s world. This paper presents innovative approach to surrogate modeling for predicting air temperature (IAT) buildings, leveraging advanced machine learning techniques. At the core of this study is application Long Short-Term Memory (LSTM) networks time-series modeling, which significantly enhances capture temporal dependencies predictions. The proposed LSTM with RWCV (Rolling Window Cross-Validation) offers significant advantages over a usual tasks, particularly due its ability adapt new data trends through rolling window mechanism. It provides more robust generalizable forecasts dynamic environments, prevents overfitting dropout cross-validation, improves model evaluation integrity. In contrast, traditional models better suited static, non-evolving datasets may not handle effectively. To rigorously assess performance, comprehensive framework developed, incorporating metrics such as mean square error (MSE) coefficient determination (R²). Additionally, novel cumulative analysis method introduced enabling real-time monitoring adjustment maintain predictive accuracy time. Test results demonstrate that losses on test dataset only marginally higher than those training dataset, indicating generalization capabilities. Loss values range from 0.0004709 0.02819861, depending building operating conditions. A comparative reveals Adaboost Gradient Boosting outperform linear regression, highlighting their potential achieving energy-efficient comfortable buildings. findings underscore efficacy IAT prediction point towards further research possibilities expansion optimization enhance conservation.

Язык: Английский

---

SolarFlux Predictor: A Novel Deep Learning Approach for Photovoltaic Power Forecasting in South Korea

Electronics, Год журнала: 2024, Номер 13(11), С. 2071 - 2071

Опубликована: Май 27, 2024

We present SolarFlux Predictor, a novel deep-learning model designed to revolutionize photovoltaic (PV) power forecasting in South Korea. This uses self-attention-based temporal convolutional network (TCN) process and predict PV outputs with high precision. perform meticulous data preprocessing ensure accurate normalization outlier rectification, which are vital for reliable analysis. The TCN layers crucial capturing patterns energy data; we complement them the teacher forcing technique during training phase significantly enhance sequence prediction accuracy. By optimizing hyperparameters Optuna, further improve model’s performance. Our incorporates multi-head self-attention mechanisms focus on most impactful features, thereby improving In validations against datasets from nine regions Korea, outperformed conventional methods. results indicate that is robust tool systems’ management operational efficiency can contribute Korea’s pursuit of sustainable solutions.

Язык: Английский

---

Exergy focused optimum solar panel tilt angle determination with improved hybrid model: The case of Turkey

Engineering Applications of Artificial Intelligence, Год журнала: 2025, Номер 145, С. 110220 - 110220

Опубликована: Фев. 8, 2025

Язык: Английский

---

Multi-Building Energy Forecasting Through Weather-Integrated Temporal Graph Neural Networks

Buildings, Год журнала: 2025, Номер 15(5), С. 808 - 808

Опубликована: Март 3, 2025

While existing building energy prediction methods have advanced significantly, they face fundamental challenges in simultaneously modeling complex spatial–temporal relationships between buildings and integrating dynamic weather patterns, particularly dense urban environments where interactions significantly impact consumption patterns. This study presents an deep learning system combining temporal graph neural networks with data parameters to enhance accuracy across diverse types through innovative modeling. approach integrates LSTM layers convolutional networks, trained using from 150 commercial over three years. The incorporates spatial a weighted adjacency matrix considering proximity operational similarities, while are integrated via specialized network component. Performance evaluation examined normal operations, gaps, seasonal variations. results demonstrated 3.2% mean absolute percentage error (MAPE) for 15 min predictions 4.2% MAPE 24 h forecasts. showed robust recovery, maintaining 95.8% effectiveness even 30% missing values. Seasonal analysis revealed consistent performance conditions (MAPE: 3.1–3.4%). achieved 33.3% better compared conventional methods, 75% efficiency four GPUs. These findings demonstrate the of prediction, providing valuable insights management systems planning. system’s scalability make it suitable practical applications smart sustainability.

Язык: Английский

---

A Novel Capacitive Model of Radiators for Building Dynamic Simulations

Thermo, Год журнала: 2025, Номер 5(1), С. 9 - 9

Опубликована: Март 11, 2025

This study addresses the critical challenge of performing a detailed calculation energy savings in buildings by implementing suitable actions aiming at reducing greenhouse gas emissions. Given high consumption buildings’ space heating systems, optimizing their performance is crucial for overall primary demand. Unfortunately, calculations such are often based on extremely simplified methods, neglecting dynamics emitters installed inside buildings. These approximations may lead to relevant errors estimation possible savings. In this framework, present presents novel 0-dimensional capacitive model radiator, most common emitter used residential The final scope paper integrate within TRNSYS 18simulation environment, 1-year simulation building-space system. radiator developed MATLAB 2024b and it carefully considers impact surface area, inlet temperature, flow rate performance. Moreover, dynamic heat transfer compared with one returned built-in non-capacitive available TRNSYS, showing that effect radiators leads an incorrect consumption. During winter season, system turned from 8 a.m. 4 p.m. 6 p.m., thermal underestimated roughly 20% commonly model.

Язык: Английский

---

Leveraging Machine Learning for Predictive Sustainability in Business Operations: A Classification Approach to Optimize Sustainable Resource Management

Studies in big data, Год журнала: 2025, Номер unknown, С. 671 - 682

Опубликована: Янв. 1, 2025

Язык: Английский

---

A Novel Intelligent Scheme for Building Energy Prediction Based On Machine Learning and Deep Learning Algorithms

Опубликована: Янв. 1, 2024

Язык: Английский

---

Detection and Early Warning of Duponchelia fovealis Zeller (Lepidoptera: Crambidae) Using an Automatic Monitoring System

AgriEngineering, Год журнала: 2024, Номер 6(4), С. 3785 - 3798

Опубликована: Окт. 18, 2024

In traditional pest monitoring, specimens are manually inspected, identified, and counted. These techniques can lead to poor data quality hinder effective management decisions due operational economic limitations. This study aimed develop an automatic detection early warning system using the European Pepper Moth, Duponchelia fovealis (Lepidoptera: Crambidae), as a model. A prototype water trap equipped with infrared digital camera controlled microprocessor served attraction capture device. Images captured by in laboratory were processed detect objects. Subsequently, these objects labeled, size shape features extracted. machine learning model was then trained identify number of insects present trap. The achieved 99% accuracy identifying target during validation 30% data. Finally, deployed field for result confirmation.

Язык: Английский

---

Development of a Hybrid Intelligence Algorithm to Estimate the Derivative Weight of Dawakin Tofa Clay for Heat Storage

AUIQ technical engineering science., Год журнала: 2024, Номер 1(2)

Опубликована: Дек. 16, 2024

The accurate prediction of thermogravimetric properties is critical for evaluating the suitability natural materials like Dawakin Tofa clay heat storage applications, but traditional linear models often fail to capture complex, non-linear relationships inherent in such datasets. This study develops a hybrid intelligence framework integrating Bilateral Neural Network (BNN), Kernel Support Vector Machine (KSVM), Step-Wise Linear Regression (SWLR), and Robust (RLR) predict derivative weight based on 5,030 experimentally obtained instances. Comprehensive data preprocessing, including normalization, feature selection, dataset splitting (80% training 20% testing), ensured high-quality inputs models. results demonstrated that significantly outperformed approaches, with BNN achieving coefficient determination R² 0.999, Mean Absolute Error (MAE) 0.004377, Percentage (MAPE) 9.6% testing dataset. Similarly, KSVM achieved an MAE 0.012134, MAPE 26.7%, indicating its robust predictive capabilities. In contrast, performed poorly, SWLR RLR yielding values 0.03 -0.41, respectively, unacceptably high 612% 53.5%. findings underscore limitations predicting complex behaviors highlight transformative potential advanced machine learning techniques KSVM. Furthermore, these align global sustainability efforts, SDG 7 12, by optimizing use locally available, eco-friendly energy storage. provides replicable leveraging artificial enhance material characterization, offering significant step toward developing sustainable solutions.

Язык: Английский

---