Prediction of heating energy consumption in university buildings based on artificial neural networks

Abstract

In this paper, the main objective is to predict heating energy consumption using a simple artificial neural network. For training and testing the network daily consumption for NTNU University campus Gløshaugen and mean outside temperatures were used. Training of the network was performed by using Levenberg–Marquardt (LM) feedforward backpropagation algorithm. Different indices of the prediction accuracy were calculated for training and testing. Simplified model showed that it can predict heating consumption with adequate accuracy. Creating a model of energy use helps in future building planning; it can provide useful information about most probable energy consumption for similar buildings, or predict energy use in different conditions.

Tema ovog rada je predviđanje potrošnje toplote korišćenjem pojednostavljene neuronske mreže. Za obučavanje i testiranje mreže korišćene su dnevne potrošnje toplote u univerzitetskom kampusu NTNU Gløshaugen, kao i srednje dnevne spoljne temperature. Za obučavanje mreže korišćen je Levenberg–Marquardt (LM) feedforward backpropagation algoritam (algoritam sa povratnim prostiranjem greške). Različiti pokazatelji kvaliteta predviđenja su izračunati i prikazani za obučavanje i testiranje mreže. Pokazuje se da pojednostavljen model može da predvidi potrošnju toplote sa zadovoljavajuće visokom tačnošću. Kreiranje ovakvih modela je korisno s aspekta energetskog planiranja izgradnje; pruža informacije o verovatnoj potrošnji toplote za slične zgrade, ili predviđa potrošnju pri različitim vremenskim uslovima.