Season-Independent PV Disaggregation Using Multi-Scale Net Load Temporal Feature Extraction and Weather Factor Fusion

TL;DR

The study addresses the challenge of disaggregating PV generation from net load in behind-the-meter setups by learning from weather data and observed net load. It introduces a Hierarchical Interpolation–based net-load feature extractor and a multi-head self-attention module to capture weather factor dependencies, followed by embedding fusion to predict daily PV generation $\hat{G}_d$ with a mean-squared error objective. The method's contributions include the HI-based load embedding, weather embedding via attention, and a fusion architecture validated on real Ausgrid data, achieving season-independent accuracy. Findings indicate robust PV disaggregation across seasons with low error metrics, enabling improved utility monitoring and demand response for distributed energy systems.

Abstract

With the advancement of energy Internet and energy system integration, the increasing adoption of distributed photovoltaic (PV) systems presents new challenges on smart monitoring and measurement for utility companies, particularly in separating PV generation from net electricity load. Existing methods struggle with feature extraction from net load and capturing the relevance between weather factors. This paper proposes a PV disaggregation method that integrates Hierarchical Interpolation (HI) and multi-head self-attention mechanisms. By using HI to extract net load features and multi-head self-attention to capture the complex dependencies between weather factors, the method achieves precise PV generation predictions. Simulation experiments demonstrate the effectiveness of the proposed method in real-world data, supporting improved monitoring and management of distributed energy systems.

Figures (2)

• Figure 1: The illustration of the proposed method.
• Figure 2: PV Disaggregation for Two Consecutive Days of a Specific Household