Modeling Multiscale Variable Renewable Energy and Inflow Scenarios in Very Large Regions with Nonparametric Bayesian Networks

In this paper, we propose a non-parametric Bayesian network method to generate synthetic scenarios of hourly generation for variable renewable energy(VRE) plants. The methodology consists of a non-parametric estimation of the probability distribution of VRE generation, followed by an inverse probability integral transform, in order to obtain normally distributed variables of VRE generation. Then, we build a Bayesian network based on the evaluation of the spatial correlation between variables (VRE generation and hydro inflows, but load forecast, temperature, and other types of random variables could also be used with the proposed framework), to generate future synthetic scenarios while keeping the historical spatial correlation structure. Finally, we present a real-life case study, that uses real data from the Brazilian power system, to show the improvements that the present methodology allows for real-life studies.