This paper presents a data-driven stochastic framework for modeling electricity consumption profiles. The motivation arises from the need for realistic load simulations to support planning and performance evaluation in small Renewable Energy Communities (RECs), where individual consumption patterns, especially those of prosumers, directly impact shared energy dynamics. Unlike conventional approaches that assume stationary seasonal distributions or rely on opaque resampling techniques, the proposed method leverages change-point detection based on Shannon entropy to identify behaviorally consistent segments in historical time series. These segments inform the training of a temporally aware Bayesian Network for day-type generation, while hour, day type, and period-specific Markov chains capture consumption transitions. The framework is applied to a real-world REC in Veneto, Italy, composed of one small-to-medium enterprise (SME) and two residential users. The SME acts as a prosumer with a 21.2 kWp photovoltaic system. Starting from one year of hourly data per user, the model generates 1,000 synthetic load profiles and simulates community behavior using Monte Carlo analysis. Results show that accounting for consumption uncertainty significantly increases the variance in monthly and yearly energy balances, nearly doubling the uncertainty compared to simulations using only stochastic production.