Scholarlink

<p>Accurate Short-Term Load Forecasting (STLF) is critical for efficient power system management, yet traditional statistical models often struggle with the non-linear and volatile nature of electricity demand. This paper addresses this issue by proposing and evaluating novel hybrid Machine Learning (ML) models,&#xa0;specifically Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), CNN-Light Gradient Boosted Machine (CNN-LightGBM), and CNN-LSTM-LightGBM,&#xa0;to enhance STLF accuracy. The methodology involves optimizing model hyperparameters using Particle Swarm Optimization (PSO) and validating performance on three diverse datasets: Saudi Electricity Company (SEC), Independent System Operator - New England (ISO-NE), and the Electric Reliability Council of Texas (ERCOT). The models, which utilize historical load, meteorological, and cyclically encoded calendar data as input features, are benchmarked against the traditional Autoregressive Integrated Moving Average with Exogenous Variable (ARIMAX) model. The findings demonstrate the significant superiority of the ML models, which achieved a Mean Absolute Percentage Error (MAPE) reduction of up to 83.03% compared to the ARIMAX model. Notably, the Random Forest (RF) model yielded the best overall performance with an MAPE of 1.379% on the SEC dataset, outperforming state-of-the-art models from recent literature. This research confirms that hybrid ML architectures, combined with rigorous feature engineering and hyperparameter optimization, provide a robust and generalizable solution for STLF. The enhanced accuracy offers substantial benefits for grid reliability, operational planning, and economic efficiency in the power sector.</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

Hybrid machine learning approach for short-term load forecasting in smart grids

  • 期刊论文

  • 发表日期：

    2026年7月13日

查看全文
  • A. Y. Alhussain,
  • S. M. Suhail Hussain

摘要

Accurate Short-Term Load Forecasting (STLF) is critical for efficient power system management, yet traditional statistical models often struggle with the non-linear and volatile nature of electricity demand. This paper addresses this issue by proposing and evaluating novel hybrid Machine Learning (ML) models, specifically Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), CNN-Light Gradient Boosted Machine (CNN-LightGBM), and CNN-LSTM-LightGBM, to enhance STLF accuracy. The methodology involves optimizing model hyperparameters using Particle Swarm Optimization (PSO) and validating performance on three diverse datasets: Saudi Electricity Company (SEC), Independent System Operator - New England (ISO-NE), and the Electric Reliability Council of Texas (ERCOT). The models, which utilize historical load, meteorological, and cyclically encoded calendar data as input features, are benchmarked against the traditional Autoregressive Integrated Moving Average with Exogenous Variable (ARIMAX) model. The findings demonstrate the significant superiority of the ML models, which achieved a Mean Absolute Percentage Error (MAPE) reduction of up to 83.03% compared to the ARIMAX model. Notably, the Random Forest (RF) model yielded the best overall performance with an MAPE of 1.379% on the SEC dataset, outperforming state-of-the-art models from recent literature. This research confirms that hybrid ML architectures, combined with rigorous feature engineering and hyperparameter optimization, provide a robust and generalizable solution for STLF. The enhanced accuracy offers substantial benefits for grid reliability, operational planning, and economic efficiency in the power sector.