Scholarlink

<p>This study presents a novel approach employing stacked ensemble of different machine learning and deep learning models for hourly plant gross load (PGL) prediction in a coal-fired thermal power plant considering python-spyder environment. The ensemble approach includes Long short-term memory (LSTM) network, Extreme gradient boosting (XG), Adaptive boosting (AD) as base learners and Random forest (RF) model as meta-learner for all the base learners. Feature selection was carried out using the XG technique integrated with the Optuna framework. Subsequently, the model hyperparameters were optimized using the same Optuna technique to enhance performance. The ensemble models were evaluated using 20% held-out test samples, applying 20-Fold Time-series cross-validation (TSCV). Model performance was assessed using a set of different evaluation metrics to ensure a comprehensive evaluation. The results confirmed that all three hybrid ensemble models achieved very similar prediction accuracy, with the XG and RF ensemble demonstrating slightly superior performance.</p> Graphical abstract <p></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

Advanced stacked ensembles for coal-fired thermal plants gross load prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月9日

查看全文
  • Ashwani Kharola,
  • Nitin Kumar,
  • Vivek John,
  • Ajay Kumar,
  • Harvinder Singh,
  • Jeewan Singh,
  • Purnendu Shekar Pandey,
  • Yohannes Mengist

摘要

This study presents a novel approach employing stacked ensemble of different machine learning and deep learning models for hourly plant gross load (PGL) prediction in a coal-fired thermal power plant considering python-spyder environment. The ensemble approach includes Long short-term memory (LSTM) network, Extreme gradient boosting (XG), Adaptive boosting (AD) as base learners and Random forest (RF) model as meta-learner for all the base learners. Feature selection was carried out using the XG technique integrated with the Optuna framework. Subsequently, the model hyperparameters were optimized using the same Optuna technique to enhance performance. The ensemble models were evaluated using 20% held-out test samples, applying 20-Fold Time-series cross-validation (TSCV). Model performance was assessed using a set of different evaluation metrics to ensure a comprehensive evaluation. The results confirmed that all three hybrid ensemble models achieved very similar prediction accuracy, with the XG and RF ensemble demonstrating slightly superior performance.

Graphical abstract