All news
Digital Twin

Initialization Is Key in Federated Short-Term Load Forecasting

US & West ·

The study focuses on federated learning for short-term load forecasting (STLF), addressing data privacy concerns. The authors identify structured heterogeneity in clients' load data: different responses to exogenous factors and distinct temporal load profiles, which degrade forecasting performance in federated learning. To mitigate these issues, they propose two model initialization strategies — global and local — that improve forecasting accuracy.

Why it matters

Relevant for forecasting tasks in the energy sector where distributed data and privacy must be considered, and adaptation to customer heterogeneity is needed.

Relevant to DT products

Original headline
Initialization Is Critical: Advancing Federated Short-Term Load Forecasting under Load Heterogeneity via Model Initialization
Read the original: arXiv — Systems and Control (eess.SY)

Translation and summary are machine-generated from the source. The full article is not republished; its rights belong to the source.

Related

ModellingUS & West

Bayesian inference for agent-based models with 90% fewer simulations

The DF-NPE method combines learned time-series summaries with two conditional normalizing flows for synthetic-data generation and posterior estimation. It supplements direct simulations with synthetic training pairs and supports pooled and ordered training strategies. In two settings of the Brock-Hommes asset-pricing model it cuts direct simulations by 90% while achieving lower posterior discrepancy and runtime.

DT.ForecastarXiv — Multiagent Systems (cs.MA) ↗
ModellingUS & West

Adapting neural posterior estimators for agent-based model inference

Agent-based models in economics and social sciences represent heterogeneous behavior and interactions through simulations, while their likelihood is often intractable for parameter estimation. The work examines likelihood-free neural inference, notably neural posterior estimation with normalizing flows providing amortized Bayesian inference. It notes that a pre-trained estimator may become inaccurate when the target task changes and discusses adapting such estimators.

DT.ForecastarXiv — Multiagent Systems (cs.MA) ↗