Enhancing Electricity Demand and Solar PV Forecasting Using Normalization and Feature Engineering in Machine Learning Models

Power systems developers need to predict solar photovoltaic (PV) power output and electricity usage with exactness to achieve energy control and maintain grid stability in modern power system development. The current leading method for energy forecasting uses machine learning models, but researchers need to conduct extensive studies to determine how different preprocessing methods and feature selection techniques affect their predictive accuracy. The research evaluates five machine learning algorithms, which include Linear Regression, Decision Tree, and Random Forest, LightGBM, and XGBoost for their ability to predict PV output and electricity demand in short and medium time periods. The research evaluates two methods that improve prediction accuracy by transforming weather data and applying data scaling techniques. The "sensed" temperature, which shows how people experience weather conditions, and the solar index ratio, which shows variations in solar radiation, are essential weather elements. The two systems monitor essential environmental factors affecting PV system performance and human activities related to electricity use. The results demonstrate that linear models achieve their best performance when data are normalized, yielding accuracy gains of 35% or more compared to unprocessed data. The performance of Random Forest and Decision Tree-based models remains consistent across different preprocessing methods because these models are robust to normalization. The ensemble models LightGBM and XGBoost achieve moderate performance enhancements through their combination with advanced feature engineering techniques. The research results demonstrate that the ML architecture requires its own feature design and preprocessing methods, which should be selected based on the specific architecture. The research establishes a dependable path to sustainable energy infrastructure by developing a repeatable machine learning system for energy prediction that delivers useful insights to enhance smart grid forecasting platforms.