Improving accuracy and convergence of federated learning edge computing methods for generalized DER forecasting applications in power grid

TL;DR

This paper proposes advanced federated learning techniques tailored for power grid DER forecasting, focusing on improved accuracy, faster convergence, and reduced communication, by leveraging hierarchical clustering, specialized models, and domain knowledge.

Contribution

It introduces novel federated learning frameworks incorporating hierarchical clustering and domain-specific insights for better DER forecasting in power grids.

Findings

Enhanced FL methods with improved accuracy

Faster convergence in non-IID data scenarios

Reduced communication overhead

Abstract

This proposal aims to develop more accurate federated learning (FL) methods with faster convergence properties and lower communication requirements, specifically for forecasting distributed energy resources (DER) such as renewables, energy storage, and loads in modern, low-carbon power grids. This will be achieved by (i) leveraging recently developed extensions of FL such as hierarchical and iterative clustering to improve performance with non-IID data, (ii) experimenting with different types of FL global models well-suited to time-series data, and (iii) incorporating domain-specific knowledge from power systems to build more general FL frameworks and architectures that can be applied to diverse types of DERs beyond just load forecasting, and with heterogeneous clients.