Scholarlink

<p>Brazil stands out globally for its high dependence on renewable energy sources within its power grid, which accounted for approximately 82% of its electricity generation in 2021, with hydroelectric power alone contributing 60.2%. Since the early 2000&#xa0;s, the Brazilian government has undertaken efforts to diversify the country’s energy matrix and reduce reliance on hydropower. Programs such as PROINFA have played a key role in supporting the development of renewable energy plants. As a result, Brazil’s installed wind power capacity has grown substantially, from 8,725.88 MW in 2015 to 25,631.7 MW in 2022. The National Energy Plan 2050 projects that wind capacity could exceed 110 GW by mid-century. Efficient operation of the National Interconnected System depends on accurate forecasting of wind patterns. The analysis compares Deep Learning architectures, including Convolutional Neural Networks, Long Short-Term Memory, a hybrid Convolutional Autoencoder–Long Short-Term Memory, against classical state-of-the-art Machine Learning methods commonly used in regression tasks, Support Vector Regression, K-Nearest Neighbors, Multilayer Perceptron, Extreme Gradient Boosting (XGBoost), and Ridge Regression (Ridge). Models are trained on wind regime data from Bahia (Brazil), with hyperparameter optimization performed via Bayesian methods. Results indicate that the CNN-ALSTM architecture achieves superior predictive performance relative to the benchmark models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Optimizing wind power forecasting with hybrid neural networks: insights from Brazil

  • 期刊论文

  • 发表日期：

    2025年7月7日

查看全文
  • José Péricles Freire,
  • Lihki Rubio,
  • Carlos E. Velasquez

摘要

Brazil stands out globally for its high dependence on renewable energy sources within its power grid, which accounted for approximately 82% of its electricity generation in 2021, with hydroelectric power alone contributing 60.2%. Since the early 2000 s, the Brazilian government has undertaken efforts to diversify the country’s energy matrix and reduce reliance on hydropower. Programs such as PROINFA have played a key role in supporting the development of renewable energy plants. As a result, Brazil’s installed wind power capacity has grown substantially, from 8,725.88 MW in 2015 to 25,631.7 MW in 2022. The National Energy Plan 2050 projects that wind capacity could exceed 110 GW by mid-century. Efficient operation of the National Interconnected System depends on accurate forecasting of wind patterns. The analysis compares Deep Learning architectures, including Convolutional Neural Networks, Long Short-Term Memory, a hybrid Convolutional Autoencoder–Long Short-Term Memory, against classical state-of-the-art Machine Learning methods commonly used in regression tasks, Support Vector Regression, K-Nearest Neighbors, Multilayer Perceptron, Extreme Gradient Boosting (XGBoost), and Ridge Regression (Ridge). Models are trained on wind regime data from Bahia (Brazil), with hyperparameter optimization performed via Bayesian methods. Results indicate that the CNN-ALSTM architecture achieves superior predictive performance relative to the benchmark models.