Generating Joint Scenarios for Renewable Generation: The Case for Non-Gaussian Models With Time-Varying Parameters

The development of medium-/long-term studies for power-system operation and planning under the uncertainty of renewable generation is a key challenge faced by power-system agents worldwide. There is a vast literature on stochastic optimization models devoted to addressing the relevant issues on both operation and planning applications. Notwithstanding, few papers focus on addressing the gaps within the subject of joint scenario generation despite the high sensibility of stochastic optimization models with regard to their input scenarios. Characterizing wind power generation (WPG) stochastic processes to devise time- and spatial-dependent scenarios, based on simulation procedures, for time horizons of one to a few years is a difficult task. Multiple regimes and non-Gaussian distributions are two of the main issues that significantly change the risk described through generated scenarios. In this paper, a new methodology to simulate long-term joint scenarios for multivariate WPG time series is presented. The proposed framework, known as generalized auto regressive score models, is derived based on a new class of time-series model with time-varying parameters and an arbitrary non-Gaussian distribution. Our case study shows, based on real data from the Brazilian power system, that the proposed methodology is capable of producing scenarios with coherent temporal and spatial dependence that are needed in power system studies.