Machine Learning Based Optimization of a Power System

Abstract

This paper presents a machine learning-driven approach for Distributed Generation optimization applied to the department of mechatronics engineering of Ahlul Bayt International University as a case study. Particular focus is on Random Forest machine learning algorithms for optimal sizing, load generation forecasting, and reinforcement learning for adaptive system control. The paper systematically analyzes the superiority of this machine learning method over conventional techniques in handling stochastic environments and improving overall system reliability. Through comparative assessment of current literature, this research directly addresses the core problem of traditional static approaches that cannot handle the real-time variations in department energy usage, developed a machine learning framework that adapts to these uncertainties.

While the implementation focuses on economic optimization and energy efficiency, the same machine learning foundation has also been extended to explicitly address voltage stability and power losses by incorporating electrical constraints into the optimization function. This has been achieved through a 47% faster economic payback (3.6 vs 6.8 years) and 47% higher operational savings (22.4% vs 15.2%) through intelligent energy management as well as Achieving 88.2% load forecasting accuracy with Random Forest algorithms resulting in 99.5% system efficiency and 100% voltage stability.