A Functional Regression Approach for Prediction in a District-Heating System

We consider the problem of short-term peak demand forecasting in a district heating system. Our dataset consists of four separated periods, with 198 days each period and 24 hourly observations within each day relative to heat consumption and climate. We take advantage of the functional nature of the data and we propose a forecasting methodology based on functional regression. The influence of exogenous explanatory variables is modelled in a suitable way. The out-of-sample performances of the proposed approach are evaluated.