Scholarlink

<p>Extreme scenarios involving abnormal load fluctuations pose significant challenge to accurate load forecasting. To address this challenge, a load forecasting model adapt to extreme scenarios is proposed, which considers the seasonal information of load data and captures the long-range dependencies. First, a hierarchical scenario extraction framework (HSEC) is proposed to comprehensively evaluate the reconstructed data from both local and global perspectives, and identify the extreme scenario based on data labels. Then, the causal multi-scale convolutional bidirectional scalar long and short-term memory network (CMSC-BisLSTM) is proposed, integrating the advantages of multiple models to improve forecasting accuracy and robustness. Finally, the cycle-aware fine-tuning (CAFT) optimization framework fine-tunes the proposed model parameters using similar-day data. Comprehensive experimental validation with two different datasets revealed that the proposed method is superior 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

Ultra-short-term load forecasting based on multi-scale optimization and extreme scenario identification

  • 期刊论文

  • 发表日期：

    2026年1月7日

查看全文
  • Jiawen Li,
  • Sen Wang,
  • Fan Sheng,
  • Yonghui Sun

摘要

Extreme scenarios involving abnormal load fluctuations pose significant challenge to accurate load forecasting. To address this challenge, a load forecasting model adapt to extreme scenarios is proposed, which considers the seasonal information of load data and captures the long-range dependencies. First, a hierarchical scenario extraction framework (HSEC) is proposed to comprehensively evaluate the reconstructed data from both local and global perspectives, and identify the extreme scenario based on data labels. Then, the causal multi-scale convolutional bidirectional scalar long and short-term memory network (CMSC-BisLSTM) is proposed, integrating the advantages of multiple models to improve forecasting accuracy and robustness. Finally, the cycle-aware fine-tuning (CAFT) optimization framework fine-tunes the proposed model parameters using similar-day data. Comprehensive experimental validation with two different datasets revealed that the proposed method is superior to the benchmark models.