Non-parametric regression modeling for stochastic optimization of power grid load forecast

This paper develops a method for building non-parametric stochastic models of multivariate distributions from large data sets. The motivation is stochastic optimization based on time series forecasting models. The proposed non-parametric stochastic modeling approach is based on multiple quantile regressions with inter-quantile smoothing. The models are built using ADMM optimization approach scalable to large datasets. As an application example, the paper considers forecasting of the loads in the electrical power grid. The forecasted load is used for the electricity procurement in the day-ahead power market. The stochastic optimization trades the costs of advance and spot procurements of the electricity. This problem is currently important because the random variability in the grid power load increases with integration of renewable generation.