XGBoost based enhanced predictive model for handling missing input parameters: A case study on gas turbine

Abstract

This work extensively develops and evaluates an XGBoost model for predictive analysis of gas turbine performance. The goal is to construct a robust prediction model by utilizing previous operational data, such as environmental variables and operational parameters. This study examines building a predictive model using the XGBoost algorithm, an ensemble learning approach known for handling huge datasets and producing robust predictions. The model is built to anticipate the gas turbine's Energy Yield (EY) output, optimize energy production efficiency, improve maintenance schedules, and enable operational decision-making within the power plant. The performance of the XGBoost model is carefully evaluated using effective evaluation metrics like RMSE, MSE, MAE, and R2 and validation methodologies, providing insights into its accuracy, robustness, and generalizability. The enhanced XGBoost predictive model's most notable feature is its ability to smoothly expand its forecasting skills to anticipate EY by applying the data acquired by anticipating positive and negative factors. Notably, the model demonstrated versatility by filling missing values in the dataset with Compressor discharge pressure (CDP) and Outlet Temperature (OT) predictions. The proposed framework intends to illustrate the practical application of predictive analytics in the sustainable energy/oil and gas industry to give significant insights into the variables influencing energy output and its potential for boosting operational efficiency and cost-effectiveness in power generation. The proposed research establishes the framework for real-time integration of the developed XGBoost model with gas turbine control systems.

Keywords

• Gas turbine
• XGBoost
• Prediction
• Energy yield
• Missing data
• Analysis