Using Nonlinear Optimization Solvers to Improve PAR(p) Coefficients Estimation in Synthetic Inflow Scenarios Generation

In this paper, we showed that using the nonlinear optimization to estimate the PAR(p) model coefficients can, sometimes, improve adherence to historical data when generating synthetic inflows, which is suit for the Stochastic Dual Dynamic Programming implemented by NEWAVE in the Brazilian energy operation planning. The used solvers were IPOPT and the Python toolbox SciPy. The accuracy of the model was checked by generating inflows via Monte Carlo simulation and outperformed, in some cases, the well established Box-Jenkins algorithm for obtaining these coefficients.