Benefits of using machine learning and probabilistic forecasting for Public Safety Power Shutoff planning

Successful execution of Public Safety Power Shutoff (PSPS) mitigations requires accurate, fine-scale weather forecasting capabilities. Several utilities in the Western US have invested in state-of-the-art high-resolution weather forecasting systems tailored specifically to predict weather and fuels conditions in their local territories. High-resolution weather forecasts are often derived from numerical weather prediction models such as the Weather Research and Forecasting (WRF) model with more sophisticated forecast systems adopting an ensemble approach whereby multiple simulations are run to sample possible forecast uncertainties. Despite advances in computational power and improved model physics, current WRF model configurations are limited to approximately 1-km grid length or larger due to the so-called model grey-zone challenge. Such limitations result in systematic forecast bias in areas of complex terrain due to inadequate terrain representation. Other biases also result from incorrect or incomplete initial conditions or model physics. Meteorologists must account for these sources of error to successfully plan for PSPS. Southern California Edison has invested heavily in the use of machine learning training algorithms and probabilistic forecasting techniques derived using observations from its expanding weather station network to overcome these challenges in the PSPS decision-making process. This presentation will overview some of the benefits achieved using machine learning and probabilistic forecasting for PSPS planning.