Energy Injection Identification enabled Disaggregation with Deep Multi-Task Learning

TL;DR

DualNILM addresses NILM in the presence of behind-the-meter energy injection by introducing a deep multi-task Transformer-based framework that jointly performs appliance state recognition and energy-injection identification. The authors formalize the injection-affected signal model, demonstrate how explicit injection modeling converts the problem from a constrained nonnegative factorization to a more tractable dual-task learning problem, and implement a CNN+Transformer architecture with Seq2Pt and Seq2Seq heads plus cross-task injection filtering. They validate DualNILM on real laboratory data with controlled injection and on PV-augmented REDD/UK-DALE datasets, showing state-of-the-art performance across both tasks and across diverse scenarios, including challenging cloud-driven PV fluctuations. The work also provides open-source PV-augmented NILM datasets to foster reproducibility and further research in renewable-integrated NILM systems.

Abstract

Non-Intrusive Load Monitoring (NILM) offers a cost-effective method to obtain fine-grained appliance-level energy consumption in smart homes and building applications. However, the increasing adoption of behind-the-meter (BTM) energy sources such as solar panels and battery storage poses new challenges for conventional NILM methods that rely solely on at-the-meter data. The energy injected from the BTM sources can obscure the power signatures of individual appliances, leading to a significant decrease in NILM performance. To address this challenge, we present DualNILM, a deep multi-task learning framework designed for the dual tasks of appliance state recognition and injected energy identification. Using a Transformer-based architecture that integrates sequence-to-point and sequence-to-sequence strategies, DualNILM effectively captures multiscale temporal dependencies in the aggregate power consumption patterns, allowing for accurate appliance state recognition and energy injection identification. Extensive evaluation on self-collected and synthesized datasets demonstrates that DualNILM maintains an excellent performance for dual tasks in NILM, much outperforming conventional methods. Our work underscores the framework's potential for robust energy disaggregation in modern energy systems with renewable penetration. Synthetic photovoltaic augmented datasets with realistic injection simulation methodology are open-sourced at https://github.com/MathAdventurer/PV-Augmented-NILM-Datasets.

Figures (8)

• Figure 1: Laboratory measurements of aggregate power demonstrate the impact of controllable micro-inverter energy injection (0-500W adjustable output) on appliance signature visibility. The blue line represents baseline aggregate power without injection, orange shows the aggregate signal under controlled laboratory injection conditions. Bottom panels display ground-truth appliance ON/OFF states, illustrating how BTM energy injection can fundamentally obscure appliance state detection from the aggregate power signal.
• Figure 2: Illustration of our proposed DualNILM.
• Figure 3: Simulated Photovoltaic energy injection on UKDALE House 1 on Apr. 05, 2013, and the corresponding Fridge and Microwave's ON/OFF states and their predictions from DualNILM and benchmarks.
• Figure 4: Visualization of Bulb I and Micro-inverter State's Ground Truth and Predictions across All Benchmark Methods on Dec.8, 2023 of our laboratory data.
• Figure 5: Comparison of overall average F1-scores for appliance state recognition across Laboratory, REDD, and UKDALE datasets.