Scholarlink

<p>This research proposes new electricity forecasting integration with Bayesian Optimized Attention-Dilated LSTM and Savitzky-Golay. Its hybrid model is aimed at enhancing the accuracy of electricity generation and consumption forecasting in developing areas. Detailed preprocessing techniques such as handling missing values through interpolation, removing outliers via Z-scores, normalization, and noise reduction with Savitzky-Golay filters significantly improved forecasting accuracy. After implementing data preprocessing using the Bayesian optimized Savitzky-Golay filter, several deep learning models such as LSTM and attention-dilated Bi-LSTM are trained. The developed model showed notable improvements over state-of-the-art techniques, yielding sMAPE values of 2.4% and 2.8% for production and consumption respectively, alongside nRMSE values of 3.1% and 3.6%. These outcomes prove the model’s adaptability to real world data changes and its practicality for improving energy management and planning. This research will assist in developing newer forecasting models in regions facing similar issues pertaining to energy consumption.</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

Enhancing Short-Term Electricity Forecasting with Advanced Machine Learning Techniques

  • 期刊论文

  • 发表日期：

    2025年9月4日

查看全文
  • Di Cui,
  • Ambe Harrison,
  • Emmanuel Fendzi-Donfack,
  • Idriss Dagal,
  • Pradeep Jangir,
  • Mohamed Metwally Mahmoud,
  • Pankaj Malkani,
  • Wulfran Fendzi Mbasso,
  • Pierre Tiako,
  • Aseel Smerat

摘要

This research proposes new electricity forecasting integration with Bayesian Optimized Attention-Dilated LSTM and Savitzky-Golay. Its hybrid model is aimed at enhancing the accuracy of electricity generation and consumption forecasting in developing areas. Detailed preprocessing techniques such as handling missing values through interpolation, removing outliers via Z-scores, normalization, and noise reduction with Savitzky-Golay filters significantly improved forecasting accuracy. After implementing data preprocessing using the Bayesian optimized Savitzky-Golay filter, several deep learning models such as LSTM and attention-dilated Bi-LSTM are trained. The developed model showed notable improvements over state-of-the-art techniques, yielding sMAPE values of 2.4% and 2.8% for production and consumption respectively, alongside nRMSE values of 3.1% and 3.6%. These outcomes prove the model’s adaptability to real world data changes and its practicality for improving energy management and planning. This research will assist in developing newer forecasting models in regions facing similar issues pertaining to energy consumption.