Budapest BI 2015

Revenues of natural gas providers and distributors naturally strongly depend on the amount of delivered natural gas and its price. The retail prices could be different for different customers. Typically the price depends on the volume of consumed gas. In that case it would be useful to forecast the total amount of consumed gas within each tariff class, i.e., for each price level. If the tariff classes didn't depend on the consumption, then it could be solved by standard forecasting methods. Otherwise, customers can change tariff classes over time. This generates a dependency structure between classes. We have developped an R package for Czech gas distributor RWE GasNet, s.r.o., which implements a prediction model taking into account the described dependency structure. The package is also able to create the explanatory variables for the model from regular invoicing data. The whole process of estimating model parameters and forecasting the number of customers and their total consumption for each tariff class can therefore be done within this R package with no need of extra data preprocessing. In the presentation I will briefly introduce the design of the prediction model and the main features of the package.