Using an artificial neural network model for natural gas heat combustion forecasting

Abstract

One of the parameters characterizing the quality of the gaseous fuel transported in gas pipeline network to consumers and being the basis for the classification of gaseous fuels is the heat of combustion. The main research hypothesis of this paper is the analysis of the possibility of using MLP 18-yi-1 neural network model to forecast the natural gas heat of combustion with a forecast error smaller than in case it calculates the heat of combustion based on the composition of natural gas predicted using the MLP 18-65-5 (Szoplik and Muchel, 2023). The training of the models was carried out on the basis of 8760 real data, presenting the hourly heat of natural gas combustion at one of the measurement points of this parameter in the pipeline network. The model takes into account the influence of calendar factors (month, day of the month, day of the week and hour of the day) and weather factors (ambient temperature) on the amount of heat of natural gas combustion in a given location of the gas network. Many MLP 18-yi-1 models were trained, differing in the number of neurons in the hidden layer and activation functions of neurons in the hidden and output layers.