Privacy-Preserving Federated Learning for Building Energy Forecasting: A Differential Privacy Analysis of Aggregation Strategies

Forecasting hourly electricity consumption across distributed building portfolios is critical for grid management and demand response, yet the sensitivity of granular energy data constrains centralised aggregation. Federated learning (FL) trains models in situ, but the cost of layering formal differential privacy (DP) guarantees atop FL remains underexplored for building-energy workloads. We train a CNN-LSTM forecasting model under three FL aggregation strategies (FedAvg, FedProx, FedBN) across four cumulative privacy budgets (ε ∈ {0.5, 1, 3, 6.5}, δ = 10⁻⁵) over a federation of 50 commercial buildings from the Building Data Genome Project 2. Counter-intuitively, all twelve DP-FL configurations achieve lower validation mean squared error (MSE) than the FedAvg-without-DP baseline, with improvements of 17% to 37%. We attribute this to per-sample gradient clipping in differentially private stochastic gradient descent (DP-SGD), which constrains heterogeneous client updates and acts as an implicit regulariser. Within the DP regime, FedAvg exhibits the expected monotonic privacy–utility frontier, FedBN displays weakly anti-monotonic behaviour, and FedProx (μ = 0.01) fails to converge at all but the tightest privacy budget. Aggregation-strategy choice can therefore rival or exceed the privacy budget in its effect on accuracy, and adopting DP-SGD can paradoxically improve federated generalisation.