Machine learning-based wind speed prediction using random forest: a cross-validated analysis for renewable energy applications

Turkish Journal of Engineering, Год журнала: 2025, Номер 9(3), С. 508 - 518

Опубликована: Март 8, 2025

Wind speed prediction plays a crucial role in renewable energy planning and optimization. This study presents comprehensive analysis of wind forecasting using Random Forest (RF) models. The research utilized high-resolution data collected throughout 2023 at the Bowen Abbot facility. Our methodology employed RF with cross-validation techniques to ensure model stability reliability. demonstrated robust performance across multiple evaluation metrics, achieving an average R² score 0.9155 (±0.0035) through 5-fold cross-validation. Error revealed consistent training, testing, validation sets, root mean square errors (RMSE) 0.6624 (±0.0098) m/s. Feature importance that 3-hour rolling was most influential predictor, accounting for 89.84% model's predictive power, followed by 1-hour (2.59%) (2.57%) lagged speeds. hierarchical temporal features suggests recent patterns are accurate predictions. error distribution showed approximately normal distributions slight deviations tails, particularly set (kurtosis: 5.2146). Key findings indicate maintains high accuracy different scales, absolute (MAE) averaging 0.4998 partitions its reliability operational deployment. These results demonstrate potential algorithms applications, providing valuable tool power generation management. study's contribute growing body on machine learning applications energy, offering insights into methodologies systems.

Язык: Английский

Short-term wind speed prediction model based on long short-term memory network with feature extraction

Earth Science Informatics, Год журнала: 2025, Номер 18(4)

Опубликована: Март 21, 2025

Язык: Английский

Short-Term Irradiance Prediction Based on Transformer with Inverted Functional Area Structure

Mathematics, Год журнала: 2024, Номер 12(20), С. 3213 - 3213

Опубликована: Окт. 14, 2024

Solar irradiance prediction is a crucial component in the application of photovoltaic power generation, playing vital role optimizing energy production, managing storage, and maintaining grid stability. This paper proposes an method based on functionally structured inverted transformer network, which maintains channel independence each feature model input extracts correlations between different features through Attention mechanism, enabling to effectively capture relevant information various features. After mixing completed linear network used predict sequence. A data processing tailored this designed, employs comprehensive preprocessing approach combining mutual information, multiple imputation, median filtering optimize raw dataset, enhancing overall stability accuracy project. Additionally, Dingo optimization algorithm suitable for self-tuning deep learning hyperparameters improving model’s generalization capability reducing deployment costs. The artificial intelligence (AI) proposed demonstrates superior performance compared existing common models forecasting can facilitate further applications generation systems.

Язык: Английский