Joint Probabilistic Day-Ahead Energy Forecast for Power System Operations

Increasing shares of wind and solar generation and rising electricity demand introduce uncertainty in power system operations. Improving short-term day-ahead forecasts of renewable energy generation and demand is critical for system operators to reduce the risk of forecasting errors, coordinate operations of the electricity market, and minimize the cost of maintaining the power system reliability. Here, we incorporate the joint probability distribution between electricity demand and energy generated from solar and wind sources to characterize system-level uncertainties in demand and supply. We develop a robust, scalable probabilistic forecasting methodology for generating system-level day-ahead forecasts of electricity demand and wind and solar generation based on publicly available weather forecasts. We combine four Sparse learning methods that identify relevant weather variables with four Bayesian learning methods that quantify the uncertainty in the forecast and evaluate each combination of these methods using proper scoring rules. Applying these models to the three zones of the California Independent System Operator, we find that the best model combination improves the system operator's forecast by 25.2%. Additionally, the confidence intervals from joint electricity demand and generation probabilistic forecast enable more effective allocation of operating reserve levels compared to current deterministic forecasts.