This paper illustrates a functional regression method for the peaks of heat demand short-term forecast of the district heating system in Turin, a northern Italian city, using load curves and weather variables. Our data-set refers to data with a fifteen minutes frequency relative to heat consumption and climate for the period from 15 october to 30 April foe years 2001-02 and 2002-03. a model has to be estimated on the base of the first period, and its out-of-sample performances have to be evaluated on the second period.