PV Fleet Modeling via Smooth Periodic Gaussian Copula

Mehmet G. Ogut1, Bennet Meyers1,2, Stephen P. Boyd1 1Stanford University, Stanford, CA, United States /2SLAC National Accelerator Laboratory, Menlo Park, CA, United States

We present a novel method for jointly modeling the energy generation from a fleet of photovoltaic (PV) systems. To accomplish this task, we propose a white-box method for finding a function that maps arbitrary time-series data to independent and identically distributed (i.i.d.) white noise. The proposed method—based on a novel approach for fitting a smooth, periodic Copula transform to data—handles many aspects of the data such as seasonal variation in the distribution of power output, dependencies among different PV systems, and dependencies across time. It consists of statistically interpretable steps and is scalable to many thousands of systems. The resulting probabilistic model of PV fleet output can perform robust anomaly detection, impute missing data, generate synthetic data, and make forecasts. In this paper, we explain the method and demonstrate these applications.