University campuses consume large amounts of electricity through classrooms, laboratories, hostels, libraries, and administrative buildings whose occupancy changes throughout the day. Conventional building management systems commonly use fixed schedules and centralized analytics, which limits their ability to adapt to local comfort needs and raises privacy concerns when occupant traces are collected at scale. This paper proposes an edge-assisted smart campus energy management system that combines short-term load forecasting, federated learning, and context-aware control. The proposed design trains local forecasting models inside each building and shares only model updates with a coordination server. A lightweight policy layer then adjusts lighting, ventilation, and noncritical loads according to occupancy, weather, tariff, and academic timetable signals. Simulated evaluation on a multi-building campus scenario shows a 17.6% reduction in energy consumption, a 21.3% reduction in peak demand, and stable comfort performance compared with rule-based scheduling.