Model-based prediction

Predicting the monthly energy consumption of a whole year is a tough problem given only sparse information about the consumers and their history. To overcome sparse data issues and to provide robust predictions we decided to use a model-based approach for prediction using two distinct estimations in a first step and later recombine them in a second step.In the first estimimation method we solely concentrate on extrapolated monthly data. In a first step we generate a model based a set of meters with many samples. These meters are normalized in mean and variance. Afterwards, we simple perform a median operation over the set of meters for each month, ignoring missing datapoints in all steps.