A Blind Source Separation Framework to Monitor Sectoral Power Demand from Grid-Scale Load Measurements

2025-12-17

AI Summary

At national/regional scales, fine-grained electricity consumption data disaggregated by sector—residential, commercial, and industrial—remain largely inaccessible, rendering the underlying electricity demand structure “invisible.” To address this, we propose a physics-constrained non-negative matrix factorization (NMF) framework for blind source separation, enabling high-fidelity dynamic decomposition of multi-sectoral electricity demand solely from aggregated, hourly transmission-level load curves. Our method integrates sector-specific load feature modeling with time-series priors to ensure physically interpretable and temporally consistent decompositions. Applied to real-world Italian grid data (2021–2023), it achieves sub-8.2% hourly sectoral reconstruction error and markedly outperforms conventional sampling-based index methods in monthly estimates. Furthermore, the framework uncovers systematic differences across sectors in seasonal patterns, weekday/weekend behavior, and demand elasticity—delivering a scalable, non-intrusive monitoring paradigm for macro-energy governance and demand-side management.

Technology Category

• Machine Learning: Matrix & Tensor Methods
• Search and Optimization: Non-convex Optimization
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Security and Privacy: Data transparency and provenance

Abstract

As we are moving towards decentralized power systems dominated by intermittent electricity generation from renewable energy sources, demand-side flexibility is becoming a critical issue. In this context, it is essential to understand the composition of electricity demand at various scales of the power grid. At the regional or national scale, there is however little visibility on the relative contributions of different consumer categories, due to the complexity and costs of collecting end-users consumption data. To address this issue, we propose a blind source separation framework based on a constrained variant of non-negative matrix factorization to monitor the consumption of residential, services and industrial sectors at high frequency from aggregate high-voltage grid load measurements. Applying the method to Italy's national load curve between 2021 and 2023, we reconstruct accurate hourly consumption profiles for each sector. Results reveal that both households and services daily consumption behaviors are driven by two distinct regimes related to the season and day type whereas industrial demand follows a single, stable daily profile. Besides, the monthly consumption estimates of each sector derived from the disaggregated load are found to closely align with sample-based indices and be more precise than forecasting approaches based on these indices for real-time monitoring.

Problem

• Monitors sectoral power demand from grid-scale load measurements
• Reconstructs hourly consumption profiles for residential, services, and industrial sectors
• Addresses lack of visibility in contributions from different consumer categories

Innovation

• Blind source separation for sectoral power demand monitoring
• Non-negative matrix factorization on grid-scale load data
• Hourly consumption profiles from aggregate high-voltage measurements