Peak-Load Forecasting Using a Functional Semi-Parametric Approach

We consider the problem of short-term peak load forecasting in a district-heating system using past heating demand data. Taking advantage of the functional nature of the data, we introduce a forecasting methodology based on functional regression approach. To avoid the limitations due to the linear specification when one uses the linear model and to the well-known dimensionality effects when one uses the full nonparametric model, we adopt a flexible semi-parametric approach based on the Projection Pursuit Regression idea. It leads to an additive decomposition which exploits the most interesting projections of the prediction variable to explain the response. The terms of such decomposition are estimated with a procedure which combines a spline approximation and the one-dimensional Nadaraya–Watson approach.