This thesis develops a decision-oriented, service-preserving techno-economic optimization framework for a multi-unit residential building in Italy. The building is modeled as 20 dwellings represented by four household archetypes to capture diversity in service-demand patterns, while the scope is limited to behind-the-meter self-consumption. An hourly input dataset for the full year 2025 is constructed, including an Italian time-of-use electricity price signal (F1/F2/F3), a gas price representation suitable for annual cost accounting, and an hourly PV generation series. Because PVGIS time series do not provide PV output for 2025, a weather-driven PV estimation layer is implemented: a supervised learning model learns the relationship between open meteorological/radiation features and PVGIS-consistent PV output on historical years, then generates the hourly PV profile for 2025 with consistent timezone and daylight-saving-time alignment.