Scholarlink

Accurate energy consumption forecasting is essential for optimizing energy management in smart grids. Traditional models often fail to capture both short-term fluctuations and long-term dependencies. To address this, we propose a hybrid CNN-LSTM model enhanced with Fourier Transform features. CNNs extract local patterns, LSTMs model temporal dependencies, and Fast Fourier Transform (FFT) incorporates frequency-domain insights to improve trend recognition. Evaluated on real-world StoreNet energy data, the proposed model outperforms baseline LSTM and traditional forecasting methods. It achieves lower Mean Squared Error (MSE) and Mean Absolute Error (MAE) while increasing the R \(^2\) score, demonstrating superior predictive accuracy. Additionally, a 5.04% SMAPE reduction highlights improved robustness. By integrating deep learning with spectral analysis, this approach provides a more accurate and stable solution for energy forecasting, supporting smart grid optimization and demand-side management.

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

Hybrid CNN-LSTM Model for Energy Consumption Prediction Using Fourier Transform Features in Renewable Energy Communities

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Mahmoud Abbasi,
  • Javier Prieto,
  • Diego Valdeolmillos,
  • Sergio Alonso Rollán

摘要

Accurate energy consumption forecasting is essential for optimizing energy management in smart grids. Traditional models often fail to capture both short-term fluctuations and long-term dependencies. To address this, we propose a hybrid CNN-LSTM model enhanced with Fourier Transform features. CNNs extract local patterns, LSTMs model temporal dependencies, and Fast Fourier Transform (FFT) incorporates frequency-domain insights to improve trend recognition. Evaluated on real-world StoreNet energy data, the proposed model outperforms baseline LSTM and traditional forecasting methods. It achieves lower Mean Squared Error (MSE) and Mean Absolute Error (MAE) while increasing the R \(^2\) score, demonstrating superior predictive accuracy. Additionally, a 5.04% SMAPE reduction highlights improved robustness. By integrating deep learning with spectral analysis, this approach provides a more accurate and stable solution for energy forecasting, supporting smart grid optimization and demand-side management.