Usage of the evolutionary designed neural network for heat demand forecast

Abstract

This paper highlights the problem of forecast model design for time series of heat demand. We propose the forecast model of heat demand based on the assumption that the course of heat demand can be described sufficiently well as a function of the outdoor temperature and the weather independent component (social components). Time of the day affects the social components. Forecast of social component is realized by means of Box-Jenkins methodology. The weather dependent component is modeled as a heating characteristic (function that describes the temperature-dependent part of heat consumption). The principal aim is to derive an explicit expression for the heating characteristics. The Neural Network Synthesis is successfully applied here to find this expression. An experiment described in the paper was realized on real life data. We have studied half-hourly heat demand data, covering four month period in concrete district heating system (DHS) from Most agglomeration and heating plant situated in Komorany, Czech Republic.