Data Efficient Solar Disaggregation with Behind-the-meter Energy Resources

  • Author / Creator
    Chen, Xinlei
  • Solar photovoltaic (PV) generation is one of the fastest-growing renewable energy sources worldwide. Almost half of this growth is projected to be behind-the-meter (BTM) installations — typically PV systems mounted on the rooftop of a home or a commercial building. Today, utilities neither have visibility into BTM distributed energy resources (DER) nor the analytical tools to reliably estimate the amount of solar power injected at any given time into their distribution feeders, a vital piece of information for crafting new tariffs and planning future investments to mitigate voltage problems and manage bidirectional power flows. This has given rise to research on signal processing and machine learning techniques that can be applied to disaggregate solar generation from the net load that is measured by a smart meter. In addition, as battery pack prices continue to decline and new pricing schemes are being introduced, many customers with PV installations will be inclined to install a battery to shift the PV output to align with the peak demand time. This "solar-plus-battery" system, when installed behind the meter, makes the disaggregation problem even harder to solve because some fluctuations in the net load that were useful for disaggregation will be smoothed out by the battery.
    This thesis aims to solve some of the challenges in the solar disaggregation problem under the assumption that the historical disaggregated data from the target home is unavailable and the deployment characteristic of BTM energy resources (e.g., PV systems and batteries) are unknown. We propose a data-efficient solar disaggregation method to estimate solar power generated by a BTM PV system and show that as long as there is one real proxy, other real proxies can be replaced with synthetic ones generated by a physical PV model with negligible loss of accuracy. We then examine how the improved solar disaggregation accuracy affects the performance of three state-of-the-art non-intrusive load monitoring (NILM) methods, namely Factorial Hidden Markov Model (FHMM), Sequence-to-Point (Seq2Point), and Denoising Autoencoder (DAE). NILM methods help residential customers understand how much money they spend on different appliances, especially energy-hungry appliances, such as air conditioner and furnace. Finally, we extend our solar disaggregation method by considering BTM battery storage. We discuss how the physical characteristics of the battery can be inferred from net meter data and how they can be used to facilitate disaggregation. Using a real dataset, we compare our methods with several state-of-the-art methods in two scenarios, which include customers with and without a BTM battery, and show that our methods outperform baselines by a clear margin.

  • Subjects / Keywords
  • Graduation date
    Spring 2022
  • Type of Item
  • Degree
    Master of Science
  • DOI
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.