Turbo compressor energy consumption modeling using machine learning techniques

Abstract

Energy consumption modeling is crucial for optimizing industrial systems, reducing operational costs, and enhancing environmental sustainability. This paper presents a comprehensive study on modeling the energy consumption of turbo compressors using advanced machine learning techniques. Initially, various regression techniques were explored to model and capture the relationships among key operating parameters. Subsequently, genetic algorithms (GA) were employed for model parameter optimization, with a particular focus on GA parameter tuning for improved convergence. Further enhancements were achieved through the Tree-structured Parzen Estimator (TPE), enabling automated hyperparameter optimization of ensemble learning methods, including Random Forest, Gradient Boosting, and XGBoost. To address heterogeneity in operating conditions, clustering was integrated into the modeling framework, facilitating the development of localized predictive models. Comparative analyses on five datasets collected under diverse environmental conditions demonstrate that ensemble methods and clustering-based approaches yield robust and highly accurate predictions. Additionally, feature importance analysis provides actionable insights for operational optimization. These findings not only advance methodologies in industrial energy modeling but also highlight their significance in achieving efficient and sustainable turbo compressor operations.