Integrating Forecasting Models Within Steady-State Analysis and Optimization

Abstract

Extreme weather variations and the increasing unpredictability of load behavior make it difficult to determine power grid dispatches that are robust to uncertainties. While machine learning (ML) methods have improved the ability to model uncertainty caused by loads and renewables, accurately integrating these forecasts and their sensitivities into steady-state analyses and decision-making strategies remains an open challenge. Toward this goal, we present a generalized methodology that seamlessly embeds ML-based forecasting engines within physics-based power flow and grid optimization tools. By coupling physics-based grid modeling with black-box ML methods, we accurately capture the behavior and sensitivity of loads and weather events by directly integrating the inputs and outputs of trained ML forecasting models into the numerical methods of power flow and grid optimization. Without fitting surrogate load models, our approach obtains the sensitivities directly from data to accurately predict the response of forecasted devices to changes in the grid. Our approach combines the sensitivities of forecasted devices attained via backpropagation and the sensitivities of physics-defined grid devices. We demonstrate the efficacy of our method by showcasing improvements in sensitivity calculations and leveraging them to design a robust power dispatch that improves grid reliability under stochastic weather events. Our approach enables the computation of system sensitivities to exogenous factors which supports broader analyses that improve grid reliability in the presence of load variability and extreme weather conditions.