The transition to electric vehicles (EVs) presents challenges and opportunities for the management of electrical networks. This paper focuses on developing and evaluating probabilistic forecasting algorithms to understand and predict EV charging behaviours, crucial for optimising grid operations and ensuring a balance between electricity demand and generation. Several forecasting approaches tailored to different time horizons are proposed across diverse model classes, including direct, bottom-up, and adaptive approaches. In all approaches, the target variable can be the load curve quantiles from 0.1 to 0.9 with 0.1 increments or prediction sets with a target coverage of 80%. Direct approaches learn from past load curves using GAMLSS or QGAM methods. Bottom-up approaches predict individual charging session characteristics (arrival time, charging duration, and energy demand) with mixture models before reconstructing the load curve. Adaptive approaches correct in real-time the prediction sets issued by direct or bottom-up approaches with conformal predictions. The experiments, conducted on real-world charging session data from Palo Alto, demonstrate the effectiveness of the proposed methods with regard to different metrics, including pinball loss, empirical coverage, and RPS. Overall, the results highlight the importance of quantifying uncertainty in load forecasts and the potential of probabilistic forecasting for EV load management.
We focus on day-ahead electricity load forecasting of substations of the distribution network in France; therefore, our problem lies between the instability of a single consumption and the stability of a countrywide total demand. Moreover, we are interested in forecasting the loads of over one thousand substations; consequently, we are in the context of forecasting multiple time series. To that end, we rely on an adaptive methodology that provided excellent results at a national scale; the idea is to combine generalized additive models with state-space representations. However, the extension of this methodology to the prediction of over a thousand time series raises a computational issue. We solve it by developing a frugal variant, reducing the number of parameters estimated; we estimate the forecasting models only for a few time series and achieve transfer learning by relying on aggregation of experts. It yields a reduction of computational needs and their associated emissions. We build several variants, corresponding to different levels of parameter transfer, and we look for the best trade-off between accuracy and frugality. The selected method achieves competitive results compared to state-of-the-art individual models. Finally, we highlight the interpretability of the models, which is important for operational applications.