Uncertainty Estimation for Energy Consumption Nowcasting

In recent years nowcasting systems have been required to perform on more complex scenarios and to a better standard than ever. This work aims at studying the relation between the performance of an array of prediction models improved via data aggregation and the measurement of uncertainty and its potential shift as a consequence of such aggregation. In order to gauge the impact of this approach we propose an experimental framework in which the predictive capabilities of distinct modeling approaches in different aggregation circumstances, is assessed in conjunction with the study of model uncertainty measurements. The results show that the not only the aggregation level, but the modeling choice can have an impact in terms of uncertainty quantification, revealing different sizes of confidence intervals. These measurements represent a novel approach to the model and aggregation level selection process.