A nonparametric Bayesian model for forecasting residential solar generation

A novel probabilistic framework to study the impact of photovoltaic-battery systems on low-voltage distribution networks

“…This work extends the non-parametric Bayesian model [26] to generate net load traces that are statistically similar to historical demand and PV generation of observed customers. The observed data was collected from the Ausgrid Solar Home Electricity Data [28].…”

“…Given these shortcomings, [26] proposed a methodology for generating residential demand and solar profiles using a Markov process specific to the features of the existing smart meter data. Instead of working up from the appliance level, the method generates the synthetic profiles by clustering a set of observed profiles using a Dirichlet process, and then generating transition matrices used in the Markov process from these clusters.…”

“…The incorporation of battery scheduling in such analysis has not been addressed in the existing literature, for two reasons: (i) a large number of residential demand and PV traces is needed for the MC analysis, and (ii) incorporating the HEM problem within the MC analysis, which is impractical due to the excessive computational burden. To solve these problems, we build on our previous work on PFA [22] and Bayesian modeling [26] by combining them into a unified MC framework 1 . Specifically:…”

“…Section 6 draws conclusions. 1 Due to the space constraints we couldn't explain our previous work [22,26] in full detail. Nevertheless, the summary provided below is detailed enough to make the paper sufficiently self-contained.…”

“…The net load traces serve as the inputs to the HEM problem in Module 2 for fast battery scheduling estimation. For more details regarding the non-parametric Bayesian model, please see [26].…”