Accurate forecasting of electricity production and electricity consumption is essential for active consumers participating in a smart low-voltage grid, particularly in distributed renewable energy generation. This paper presents a forecasting approach for a photovoltaic (PV) system installed at a primary school in Bosnia and Herzegovina, focusing on predicting total electricity consumption over a four-day horizon. Given the stochastic nature of solar generation and electricity consumption patterns, the research applies Least Squares Boosting (LSBoost), a machine learning technique known for its ability to capture nonlinear dependencies and improve predictive accuracy. The forecasting model is trained using historical PV production, imported and exported electricity, total electricity consumption, and meteorological variables (temperature, wind speed, and humidity). The model’s performance is evaluated using predictive accuracy indicators, achieving high predictive accuracy, with PV production forecasts attaining an R2 of 0.937 and total consumption an R2 of 0.889. The analysis of the forecasting period (November 1st–4th, 2024) highlights the model’s ability to accurately reflect operational characteristics of the school, particularly the reduction in electricity consumption during non-operational days (weekends) and increased consumption on active school days. The results confirm that LSBoost provides reliable energy forecasts for active consumers, enabling more effective energy management, improved financial planning, and optimized grid interactions.