Repository logo

Theses and Dissertations

Permanent URI for this collection

https://hdl.handle.net/10217/199889

Browse

Browsing Theses and Dissertations by Subject "AMI data"

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    Disaggregation of net-metered advanced metering infrastructure data to estimate photovoltaic generation
    (Colorado State University. Libraries, 2019) Stainsby, Wendell Jay, author; Young, Peter, advisor; Zimmerle, Daniel, committee member; Aloise-Young, Patricia, committee member
    Advanced metering infrastructure (AMI) is a system of smart meters and data management systems that enables communication between a utility and a customer's premise, and can provide real time information about a solar array's production. Due to residential solar systems typically being configured behind-the-meter, utilities often have very little information about their energy generation. In these instances, net-metered AMI data does not provide clear insight into PV system performance. This work presents a methodology for modeling individual array and system-wide PV generation using only weather data, premise AMI data, and the approximate date of PV installation. Nearly 850 homes with installed solar in Fort Collins, Colorado, USA were modeled for up to 36 months. By matching comparable periods of time to factor out sources of variability in a building's electrical load, algorithms are used to estimate the building's consumption, allowing the previously invisible solar generation to be calculated. These modeled outputs are then compared to previously developed white-box physical models. Using this new AMI method, individual premises can be modeled to agreement with physical models within ±20%. When modeling portfolio-wide aggregation, the AMI method operates most effectively in summer months when solar generation is highest. Over 75% of all days within three years modeled are estimated to within ±20% with established methods. Advantages of the AMI model with regard to snow coverage, shading, and difficult to model factors are discussed, and next-day PV prediction using forecasted weather data is also explored. This work provides a foundation for disaggregating solar generation from AMI data, without knowing specific physical parameters of the array or using known generation for computational training.