Scholarlink

<p>Accurate forecasting of electricity consumption is crucial for optimizing energy management, reducing operational costs, and ensuring the sustainability of power systems. This study introduces a two-stage framework for short-term electricity consumption prediction by integrating dimensionality reduction techniques—Principal Component Analysis (PCA) and Independent Component Analysis (ICA)—with machine learning models, including Random Forest, Support Vector Regression (SVR), Linear Regression (LR), Artificial Neural Networks (ANN), and Long Short-Term Memory (LSTM) networks. The original 12-dimensional dataset was reduced to two key components using PCA and ICA to eliminate redundancy and noise while preserving essential features. These transformed datasets were then used to train predictive models. Model performance was evaluated using metrics such as the coefficient of determination (R<sup>2</sup>), Root Mean Square Error (RMSE), and the Willmott Index (WI). The results indicate that PCA generally outperforms ICA across most models. For instance, Random Forest with PCA achieved R<sup>2</sup> = 0.79 and RMSE = 44.79, compared to R<sup>2</sup> = 0.763 and RMSE = 48.75 using ICA. Likewise, the LSTM model performed significantly better when combined with PCA (R<sup>2</sup> = 0.78, WI = 0.83) than with ICA. Paired t test results further support these findings, showing statistically significant improvements in Mean Absolute Error (MAE) for PCA over ICA in ANN and LSTM models (<i>p</i> &lt; 0.05). The findings demonstrate that combining advanced dimensionality reduction—particularly PCA—with powerful machine learning algorithms, especially deep learning models like LSTM, leads to more robust and accurate electricity consumption forecasting. Future work could explore hybrid feature extraction methods and transformer-based architectures to enhance prediction performance across diverse datasets.</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

Prediction of electrical energy consumption using principal component analysis and independent components analysis

  • 期刊论文

  • 发表日期：

    2025年6月23日

查看全文
  • Nima Saeedi,
  • Deniz Baharvand,
  • Kimia Shirini,
  • Sina Samadi Gharehveran

摘要

Accurate forecasting of electricity consumption is crucial for optimizing energy management, reducing operational costs, and ensuring the sustainability of power systems. This study introduces a two-stage framework for short-term electricity consumption prediction by integrating dimensionality reduction techniques—Principal Component Analysis (PCA) and Independent Component Analysis (ICA)—with machine learning models, including Random Forest, Support Vector Regression (SVR), Linear Regression (LR), Artificial Neural Networks (ANN), and Long Short-Term Memory (LSTM) networks. The original 12-dimensional dataset was reduced to two key components using PCA and ICA to eliminate redundancy and noise while preserving essential features. These transformed datasets were then used to train predictive models. Model performance was evaluated using metrics such as the coefficient of determination (R2), Root Mean Square Error (RMSE), and the Willmott Index (WI). The results indicate that PCA generally outperforms ICA across most models. For instance, Random Forest with PCA achieved R2 = 0.79 and RMSE = 44.79, compared to R2 = 0.763 and RMSE = 48.75 using ICA. Likewise, the LSTM model performed significantly better when combined with PCA (R2 = 0.78, WI = 0.83) than with ICA. Paired t test results further support these findings, showing statistically significant improvements in Mean Absolute Error (MAE) for PCA over ICA in ANN and LSTM models (p < 0.05). The findings demonstrate that combining advanced dimensionality reduction—particularly PCA—with powerful machine learning algorithms, especially deep learning models like LSTM, leads to more robust and accurate electricity consumption forecasting. Future work could explore hybrid feature extraction methods and transformer-based architectures to enhance prediction performance across diverse datasets.