Air Handling Units (AHUs) substantially affect the energy demand of buildings. Their operation and energy use depend on several factors, such as indoor or outdoor temperatures and occupant activities; this requires AHUs’ performance optimization through operation data modeling and suitable control algorithms. Given multiple AHUs, if their operation is sufficiently similar, a Federated Learning (FL) model can be constructed, resulting in a global control algorithm. Here, we present the modeling of Automated Supply Air Temperature (ASAT) for three AHUs that serve multiple spaces in a university building. Clustering, Wasserstein distance, and Kullback-Leibler divergence (KLD) methods were used to identify the similarities or differences in the operation of ASAT based on historical data. Clustering and KLD indicated differing AHUs patterns, while the Wasserstein distance showed a high level of similarity. The technical reasons underlying the differences in operational principles were identified and explained. The FL developed in this study was implemented by modifying the regularization parameter, reflecting the similarities between the graph nodes.