Scholarlink

In this paper, a methodology for power prediction within a Combined Cycle Power Plant (CCPP) was developed and implemented, leveraging probabilistic regression models and ensemble techniques. The initial phase involved the selection and evaluation of probabilistic regression models, with the Gradient Boosting Regressor chosen for its ability to capture intricate dataset connections. Subsequently, a Stacking Regressor was introduced in acknowledgment of diverse patterns within the CCPP dataset. This ensemble technique integrated predictions from Random Forest, Support Vector Regressor (SVR), and Neural Network base regressors, fostering a collective understanding of dataset nuances. Expanding our investigation, the third phase introduced Random Forest as an individual probabilistic regression model, contributing to a diverse range of models considered. Ensemble techniques in subsequent phases integrated model predictions, with the Voting Regressor combining outputs from Gradient Boosting, SVR, and Random Forest to provide a holistic power forecast. In the final phases, the Bagging Regressor with Decision Trees and AdaBoost Regressor with shallow Decision Trees were applied, showcasing the potential of aggregating predictions from separate models and adapting to various dataset patterns. Model performance was systematically assessed using Mean Squared Error, aiding in the selection of models that aligned with the dataset characteristics. This methodology not only improved predictive accuracy but also acknowledged and quantified uncertainties in power projections, essential for effective energy management in dynamic operational scenarios. The results highlight the versatility and robustness of the proposed approach in the realm of power prediction within CCPPs.

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

Power Prediction in CCPP Through ML-Based Probabilistic Regression Models and Ensemble Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • P. B. V. Rajarao,
  • S. Ushanag,
  • T. Prabhakara Rao,
  • G. Jose Moses,
  • A. Lakshmanarao

摘要

In this paper, a methodology for power prediction within a Combined Cycle Power Plant (CCPP) was developed and implemented, leveraging probabilistic regression models and ensemble techniques. The initial phase involved the selection and evaluation of probabilistic regression models, with the Gradient Boosting Regressor chosen for its ability to capture intricate dataset connections. Subsequently, a Stacking Regressor was introduced in acknowledgment of diverse patterns within the CCPP dataset. This ensemble technique integrated predictions from Random Forest, Support Vector Regressor (SVR), and Neural Network base regressors, fostering a collective understanding of dataset nuances. Expanding our investigation, the third phase introduced Random Forest as an individual probabilistic regression model, contributing to a diverse range of models considered. Ensemble techniques in subsequent phases integrated model predictions, with the Voting Regressor combining outputs from Gradient Boosting, SVR, and Random Forest to provide a holistic power forecast. In the final phases, the Bagging Regressor with Decision Trees and AdaBoost Regressor with shallow Decision Trees were applied, showcasing the potential of aggregating predictions from separate models and adapting to various dataset patterns. Model performance was systematically assessed using Mean Squared Error, aiding in the selection of models that aligned with the dataset characteristics. This methodology not only improved predictive accuracy but also acknowledged and quantified uncertainties in power projections, essential for effective energy management in dynamic operational scenarios. The results highlight the versatility and robustness of the proposed approach in the realm of power prediction within CCPPs.