Predicting Offshore Oil Slick Formation: A Machine Learning Approach Integrating Meteoceanographic Variables

Water, Год журнала: 2025, Номер 17(7), С. 939 - 939

Опубликована: Март 24, 2025

The presence of oil slicks in the ocean presents significant environmental and regulatory challenges for offshore processing operations. During primary oil–water separation, produced water is discharged into ocean, carrying residual oil, which measured using total grease (TOG) method. formation spread are influenced by metoceanographic variables, including wind direction (WD), speed (WS), current (CD), (CS), wave (WWD), peak period (PP). In Brazil, limits impose sanctions on companies when exceed 500 m length, making accurate prediction their occurrence extent crucial operators. This study follows three main stages. First, performance five machine learning classification algorithms evaluated, selecting most efficient method based metrics from a Brazilian company’s slick database. Second, best-performing model used to analyze influence variables TOG levels detection probability. Finally, third stage examines detected identify key contributing factors. results enhance decision-support frameworks, improving monitoring mitigation strategies discharges.

Язык: Английский

---

Improving SMART Learning: Course Completion via AI-Driven Hybrid System Integration in Big Data

Telematics and Informatics Reports, Год журнала: 2025, Номер unknown, С. 100199 - 100199

Опубликована: Апрель 1, 2025

Язык: Английский

---

Solar radiation prediction: A multi-model machine learning and deep learning approach

AIP Advances, Год журнала: 2025, Номер 15(5)

Опубликована: Май 1, 2025

The increasing integration of renewable energies into electrical grids necessitates accurate forecasting meteorological variables, particularly solar irradiance. This study presents a novel long-term irradiance approach, utilizing data from the National Renewable Energy Laboratory spanning 1988–2022. Focusing on five input variables—solar irradiance, dew point, temperature, relative humidity, and wind speed—this evaluates predictive performance 13 data-driven models, comprising ten machine learning (ML) three deep (DL) algorithms. Among them, gradient boosting regressor (GBR) recurrent neural network (RNN) emerged as top performers in ML learning, respectively. In order to choose most suitable model for long short term, four forecast time-horizons (1, 8, 16, 24 h) were also taken consideration models. A feature selection process using Pearson’s coefficient identified relevant inputs, while quantile regression was employed uncertainty assessment, mean prediction interval, interval coverage probability demonstrates that RNN excels short-term predictions, GBR is more effective forecasts. new hybrid approach GBR-RNN developed, achieving superior terms RMSE, MAE, R2 metrics. multi-model integrating both DL techniques, enhances by addressing considering various horizons. findings contribute ongoing advancement energy providing robust, accurate, uncertainty-aware Moreover, this helps identify best-performing model, enabling reliable precise forecasts management. highlights improvement methods importance selecting best accuracy.

Язык: Английский

---

A New Precipitation Prediction Method Based on CEEMDAN-IWOA-BP Coupling

Water Resources Management, Год журнала: 2022, Номер 36(12), С. 4785 - 4797

Опубликована: Авг. 10, 2022

Язык: Английский

---

Surrogate model of a HVAC system for PV self-consumption maximisation

Energy Conversion and Management X, Год журнала: 2023, Номер 19, С. 100396 - 100396

Опубликована: Май 26, 2023

In the last few years, energy efficiency has become a challenge. Not only mitigating environmental impact but reducing waste can lead to financial advantages. Buildings play an important role in this: they are among biggest consumers. So, finding manners reduce consumption is way minimise waste, and technique for that creating Demand Response (DR) strategies. This paper proposes novel decrease computational effort of simulating behaviour building using surrogate models based on active learning. Before going straight problem building, which complex computationally costly, approach learning smaller problem: with reduced simulations, regress curve voltage versus current thermo-resistor. Then, implements model building. The goal be able learn pattern limited number simulations. result given by used set reference temperature, maximising PV self-consumption, usage from grid. Thanks surrogate, total time spent map all possible scenarios around 7 times.

Язык: Английский

---