Adaptive Forecasting of Extreme Electricity Load

Abstract

Electricity load forecasting is a necessary capability for power system operators and electricity market participants. Both demand and supply characteristics evolve over time. On the demand side, unexpected or extreme events as well as longer-term changes in consumption habits affect demand patterns. On the production side, the increasing penetration of intermittent power generation significantly changes the forecasting needs. We address this challenge in three ways. First, we consider probabilistic (quantile) rather than point forecasting; indeed, uncertainty quantification is required to operate electricity systems efficiently and reliably. The probabilistic forecasts are generated using both linear and non-linear quantile regressions applied to the residuals of the mean forecasting model. Second, our approach is Adaptive; we have developed models that incorporate the most recent observations to automatically respond to changes in the underlying process. Our adaptation methodology leverages the Kalman filter, which has previously been successfully employed for adaptive load forecasting, as well as Online Gradient Descent - a combination of an incremental strategy and pinball loss. Third, we extend the adaptive setting to Extreme scenarios by using the aforementioned methods to compute an adaptive threshold used as a reference in recently developed machine learning models targeting extreme values. Finally, we apply our different approaches on the french daily electricity consumption as use case.