Load Estimation of Different Types of Domestic Users Using Machine Learning Methods and Optimal Battery Sizing

Electricity, a human necessity since the 19th century, has gained ever-increasing importance as the world population grows, set to reach 10 billion by 2050. The urgency for sustainable and nature-friendly energy production, as well as efficient consumption, parallels the rising demand. Rapid urbanization and industrialization are increasing energy needs and greenhouse gas emissions, prompting countries to reduce emissions through policies and environmental protocols. This study explores the challenges of integrating variable, and weather-dependent renewable energy sources into the grid, which necessitates accurate energy consumption prediction. The load consumption of various domestic users is aimed to be predicted and battery sizing is intended to be optimized accordingly. Data from 15 households of varying sizes with 1 minute resolution, spanning over a year with minute-resolution, was used. Machine learning models, including LSTM, Random Forest Regressor, XGBoost, and Linear Regression were employed, with temperature, holidays, and sunrise/sunset times identified as significant features. The study extends beyond load prediction, promoting consumer savings through variable electricity prices and advocating for battery use for reliable electricity supply. This work represents a pioneering effort in battery optimization based on load prediction data, facilitating a balanced, economical, and sustainable power system.