Targeted Efficiency: Using Customer Meter Data to Improve Efficiency Program Outcomes
- Author(s): Borgeson, Samuel Dalton;
- Advisor(s): Callaway, Duncan;
- et al.
Energy efficiency (EE) and demand response (DR) programs are designed to reduce energy consumption, mitigate grid capacity constraints, support intermittent renewable energy integration, and reduce pollution for less than the cost of additional generation. However, the savings and flexibility achieved by EE and DR programs are contingent on programs finding and enrolling customers well matched to program objectives. Many EE programs currently rely on broad but shallow savings from prescriptive measures, but growing interest in deeper and more reliable savings is leading to increased attention for program planning and targeting using empirical criteria. Through data gathered by smart meters and software designed to manage and analyze large data sets, the tools required to cost effectively characterize, target, and change patterns of energy demand are beginning to emerge. This dissertation consists of three chapters on the analysis of meter data and one on the policy implications of these new capabilities.
Chapter 2 analyzes hourly electricity and daily natural gas smart meter readings from 30,000 residential customers of Pacific Gas and Electric (PG&E). Meter data is used to derive distributions of previously unobserved characteristics of the housing stock. We show that the targeting of EE and DR programs would be improved through selection of households based on their positions within these distributions. It follows that every utility (or public utility commission) with sufficient metering infrastructure could apply similar techniques to improve the targeting, implementation, and evaluation of their energy efficiency and demand response programs.
Chapter 3 uses regression models of patterns in daily household electricity consumption to estimate the physical and operational characteristics of homes. We apply semi-physical regressors designed to capture patterns in the space-heating and cooling, scheduling, and occupancy of homes. When applied to data from approximately 160,000 PG&E customers, this approach supports an evaluation of competing regression model formulations and provides distributions of model coefficients used to evaluate patterns of domestic energy use, including annual and system peak coincident air conditioning loads, cooling set points, day of week scheduling, and lighting energy.
Chapter 4 presents a method for estimating the hourly timing of occupant driven loads based on smart meter data. The residuals of a predictive regression model are assumed to include occupant activities because occupant controlled energy use is not fully determined by externally observable factors. Occupant activity timing is converted into empirical distributions of the probability of such events by hour-of-day or day-of-week. With estimates calculated for approximately 25,000 PG&E customers and grouped using K-means clustering, prevailing patterns are interpreted as the result of occupant lifestyles — with applications in efficiency and demand response program targeting.
Drawing upon the applications developed in the preceding chapters, Chapter 5 discusses the potential for using smart meter data to support public interest utility programs in the context of ongoing concerns over public disclosure and privacy concerns, including malicious use by bad actors and inappropriate commercial use. We propose differentiated levels of access to meter data, with access to data mediated by delegated analysis, which allows stakeholders to receive the outputs of approved algorithms without requiring direct access to sensitive data. Such a system would provide privacy protection and oversight without foreclosing on creative and innovative uses of meter data.