Adaptive Methods for Short-Term Electricity Load Forecasting During COVID-19 Lockdown in France

Abstract

The coronavirus disease 2019 (COVID-19) pandemic has urged many governments in the world to enforce a strict lockdown where all nonessential businesses are closed and citizens are ordered to stay at home. One of the consequences of this policy is a significant change in electricity consumption patterns. Since load forecasting models rely on calendar or meteorological information and are trained on historical data, they fail to capture the significant break caused by the lockdown and have exhibited poor performances since the beginning of the pandemic. In this paper we introduce two methods to adapt generalized additive models, alleviating the aforementioned issue. Using Kalman filters and fine-tuning allows to adapt quickly to new electricity consumption patterns without requiring exogenous information. The proposed methods are applied to forecast the electricity demand during the French lockdown period, where they demonstrate their ability to significantly reduce prediction errors compared to traditional models. Finally, expert aggregation is used to leverage the specificities of each predictions and enhance results even further.

State-Space Models for Online Post-Covid Electricity Load Forecasting Competition

“…Our work extends a previous study on the French electricity load [17] where a state-space approach was presented to adapt generalized additive models in the context of online learning. The idea is to plug a Kalman filter on the estimated effects of a GAM to gain in online reactivity.…”

“…To adapt GAM and MLP we linearize the models and x t is just another feature representation. We freeze the non-linear effects in the GAM as in [17], and x t contains the different effects, linear and nonlinear. We apply a similar approach for the MLP, for which we freeze the deepest layers and we learn the last one, that is x t is the final hidden state, see Figure 6.…”

“…However there is no global guarantee and in our case it performs poorly. We propose instead the following settings, building on [17]:…”

“…Online robust aggregation of experts [29] is a powerful model agnostic approach for time series forecasting, already applied to load forecasting during the lockdown in [17]. We use the ML-Poly algorithm proposed in [30] and implemented in the R package opera [31] to compute these online weights.…”

“…Furthermore, we introduce an expert named GAM_SAT forecasting each day with the GAM as if it were a Saturday motivated by [17].…”