Forecasting Natural Gas Consumption in Hakkari Province Using Artificial Neural Network

Abstract

Natural gas consumption is a continuous process that requires an exact supply to prevent
interruptions in domestic use, particularly under harsh climatic conditions. This study is presented to
forecast natural gas consumption in Hakkari Province using the Multiple Linear Regression (MLR) and
Artificial Neural Network (ANN) models for the first 9 months of 2023. Training data consisted of
relative humidity, sunshine duration, temperature, solar intensity, radiation, and subscriber numbers for
the 2020-2022 period. The consumption was selected as the dependent output parameter, while the other
parameters were independent input variables. Data analysis and training were performed using Python in
the Visual Studio Code (VSCode) environment. Data processing was conducted using the Min-Max
normalization method to ensure data consistency, and the 10-fold cross-validation technique was applied.
A fully connected feed-forward neural network was used in the ANN model, which consisted of three
hidden layers containing 32, 16, and 8 neurons. LeakyReLU was preferred as the activation function in
the hidden layers to provide non-linearity, while a linear activation function was implemented in the
output layer for regression. The Adaptive Moment Estimation (Adam) algorithm was selected as the
optimizer to update the network weights and minimize the loss function more efficiently. The R square,
RMSE and MAE values were found as 0.755, 0.128, 0.104 for the MLR model, and 0.943, 0.06, 0.034 for
the ANN model, respectively. The results indicate that the ANN model showed better performance in
terms of forecasting compared to the MLR model.