A Markov Chain Monte Carlo approach for the estimation of photovoltaic system parameters

Abstract

Knowledge of the installation parameters of photovoltaic systems is essential in the context of grid management: by relating these parameters to performance data, forecasting models may be optimised to improve the management of power flow into the grid. In the case of small residential systems, these parameters are often not available. We present a novel method for determining the azimuth (ϕ), tilt (θ) and rated power (P) of photovoltaic systems, using openly available data over the course of 2016–2018 of 12 photovoltaic systems in PVOutput. This method consists of two steps: firstly we identify a candidate list of clear days by computing descriptive statistics of a larger set of 80 PVOutput system profiles. In the second step we compare the observed clear-day profiles, of the aforementioned 12 systems, with modelled clear-sky profiles from the PVLib library. The fits are performed employing a Markov Chain Monte Carlo (MCMC) approach, implemented with the Emcee package: the most favoured parameters and their associated uncertainties, for any given day, are obtained by sampling from the posterior assuming a Gaussian sampling distribution. The results for our 12 systems are in good agreement with the PVOutput metadata.