Hakkari province natural gas consumption estimation with artificial neural networks

Today, energy consumption has become an area that needs to be meticulously managed, taking into account important factors such as environmental impacts and sustainability of resources. Predicting future trends in natural gas consumption and improving efficient resource management is of strategic importance in the energy sector. In this study, it is aimed to examine the relationship between relative humidity, sunshine duration, temperature, insolation intensity, radiation and subscriber numbers and the amount of natural gas consumption for the years 2020, 2021 and 2022 and to predict the amount of natural gas consumption for the first 9 months of 2023. For this purpose, multiple linear regression analysis and randomforest, gradientboosting and xgboost algorithms, which are frequently used in machine learning, were used. In the study, consumption amount was selected as the dependent variable and other data as independent variables. The data analysis and forecasting process was performed with Visual Studio Code (VSCode) integration using Python programming language. The performance of the models was evaluated based on the Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE) and R-Square (R²). According to the results obtained, the RandomForest algorithm showed the best performance. The MSE value of the predictions obtained with the RandomForest model was 165396993, RMSE value was 12860, MAPE value was 0.2% and R² value was 0.92. These results show that the RandomForest algorithm provides the most accurate and reliable results in natural gas consumption forecasts for the first 9 months of 2023. Key Words: NaturalGas, Forecasting, Machine Learning