Scholarlink

<p>The accurate forecasting of solar power generation are crucial for optimizing energy management, maintaining grid stability, and supporting renewable energy integration. This study proposes a transformer-based model for solar power prediction, using its self-attention mechanism to capturing long-term dependencies and temporal patterns in time-series data. The model’s performance is compared with four advanced neural architectures: BiLSTM, Attention-LSTM, CNN, and GRU. The Transformer model outperformed all other models, achieving the lowest Mean Absolute Error (MAE: 0.0025) and the highest <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation> value (0.9998), indicating its superior accuracy and robustness. This paper provides a detailed methodology, covering data preprocessing, model architecture, and hyperparameter tuning, along with the mathematical formulations that define each model.</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

A transformer-based approach for solar power prediction: capturing temporal patterns for grid stability

  • 期刊论文

  • 发表日期：

    2026年7月3日

查看全文
  • Dashrath Nishad,
  • Ashutosh Kumar Singh

摘要

The accurate forecasting of solar power generation are crucial for optimizing energy management, maintaining grid stability, and supporting renewable energy integration. This study proposes a transformer-based model for solar power prediction, using its self-attention mechanism to capturing long-term dependencies and temporal patterns in time-series data. The model’s performance is compared with four advanced neural architectures: BiLSTM, Attention-LSTM, CNN, and GRU. The Transformer model outperformed all other models, achieving the lowest Mean Absolute Error (MAE: 0.0025) and the highest \(R^2\) value (0.9998), indicating its superior accuracy and robustness. This paper provides a detailed methodology, covering data preprocessing, model architecture, and hyperparameter tuning, along with the mathematical formulations that define each model.