Federated learning captures static features in time series forecasting unevenly
Learning From Static Context in Federated Time Series Forecasting
Machine Learning
Summary
Predicting future values in many related time series is harder when each data owner keeps their data private. This paper looks at a popular forecasting model that uses a fixed description of each series to tell them apart. The authors find that some federated training methods can learn from these fixed descriptions almost as well as centralized training, while others struggle. The best-performing federated method matches or surpasses training models locally for each series, but the results vary with the dataset and training approach.
What this means in practice
- •For energy management teams: Improve wind power output and building energy forecasts using federated models that learn static features across multiple data owners without sharing raw data.
- •For electric utilities: Enhance load forecasting accuracy using federated learning methods that retain critical static information while respecting privacy of customer data.
Authors
Viktor Walter, Andreas Wagner, Fabian Stricker
Abstract
Time series forecasting is shifting from locally trained, per-series models to global models trained across many heterogeneous series, which requires learning what sets the series apart. When the raw data cannot be centralized, such a global model is trained with federated learning (FL), one client per data owner. State-of-the-art forecasters such as the Temporal Fusion Transformer (TFT) tell series apart through a static input that describes each series and does not change over time. In the special case of a federated setting where each client holds a single series, this static input is constant within each client and varies only across clients. We ask whether a federated TFT can learn from it. Per-batch FedAdam, which aggregates gradients after every batch, uses the static input almost as well as central training, whereas communication-efficient schemes with local epochs (FedAvg, FedAdam+Prox) extract far less from it and gain nothing significant on either application dataset. At the same local computation, an Adam-compatible SCAFFOLD keeps the static pathway responsive on wind power and gains significantly there. On building energy its error still drops, but the static pathway falls silent, so drift and communication frequency cannot be fully separated as the cause. Only per-batch FedAdam matches or beats the per-client local models across all datasets. We confirm the effect on building-energy meter data, a wind-power dataset, and a controlled synthetic dataset. The loss is not specific to the TFT: on the synthetic and the wind data two further architectures with different static-conditioning mechanisms show it as well.