Short term hourly forecasting of gas consumption using neural networks

Abstract

This paper presents a neural network based model for forecasting gas consumption .for residential and commercial consumers. A feedforward neural network with sigmoid nodes and one hidden layer was trained by backpropagation. The model was validated on real data from a distribution area covering 7% of the total consumption in Croatia, consisting mostly of residential and commercial consumers.

Forecasting peak-day consumption for year-ahead management of natural gas networks

“…In the gas industry, alternative methods can be found to model gas consumption for general forecasting and gas management purposes, including: regression models (Gas Networks Ireland, 2007); auto-regressive integrated moving-average or autoregressive models including exogenous variables (Aras and Aras, 2004;Brabec et al, 2008); neural networks (Brown et al, 1994;Kizilaslan and Karlik, 2008;Peharda et al, 2001;Khotanzad et al, 2000); and generalised additive models (Brabec et al, 2010). In relation to peak consumption estimation, regression-based models are considered to offer a more transparent methodology compared to the alternatives.…”