Supervised Machine Learning Models Evaluation to Forecast Preventive Maintenance Costs in a Thermal Power Plant

This praxis explores the application of supervised machine learning models to predict maintenance costs in a thermal power plant. It addresses the challenges faced by plant managers in achieving accurate cost predictions, mainly because traditional methods are typically limited in adaptability, and non effective for short-term forecasting, which leads to frequent budget overruns. Historical data about plant performance and maintenance activities is used to evaluate the predictive accuracy of four machine learning techniques: Time Series (ARIMA, SARIMA), Linear Regression (ElasticNet), Decision Trees (XGBoost), and Long Short-Term Memory (LSTM) networks through mean absolute percentage error (MAPE), and other performance metrics such as mean absolute error (MAE), and root mean squared error (RMSE) using a holdout data set. Results are aligned with ASTM E2516-11 cost estimate accuracy standards. According to the results, XGBoost achieved robust performance with a MAPE up to 20% followed by SARIMA at 28%. The most significant contributors to short-term (12-month) cost prediction are time-based factors (e.g., monthly cyclic trends and historical cost rolling statistics, which account for 90 to 96% of the weight). This study contributes to the maintenance management practices in the energy sector, offering a data-driven framework adaptable to similar industrial contexts. Future research may focus on model refinement using hybrid ensemble models, integration with other predictor factors, and testing cross-plant applicability and industry-wide adoption in asset-intensive plants.

Author

• RODRIGUEZ, JAVIER Hernando

Language

• English

Keyword

• XGBoost
• Power Plant
• Preventive Maintenance Costs
• Maintenance Budgeting
• Machine Learning
• Framework

Date created

• 2025

Type of Work

• Dissertation

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Engineering Management

Degree

• D.Engr.

Advisor

• Yassan, Rebecca M.

Committee Member(s)

• Fagan, Cory
• Blackford, Joseph P.

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/5138jf755