Turkey's energy demand forecast until 2040 using machine learning algorithms

The rapid increase in industrialization, coupled with population growth, has led to an increased energy demand. However, machine learning algorithms have come to the forefront to make anticipatory energy predictions to meet this emerging energy demand. Particle Swarm Optimization (PSO), Artificial Neural Networks (ANN), and Support Vector Regression (SVR) are among these algorithms. In this study, electricity demand forecasting for Turkey between 2020 and 2040 was conducted using the PSO, ANN, and SVR algorithms. To perform these prediction processes, annual electricity consumption data from 1980 to 2019 were obtained from TEIAS (Turkey Electricity Transmission Corporation), population data from TUIK (Turkish Statistical Institute), and export, import, gross domestic product (GDP) data from the World Bank open data set. PSO, ANN, and SVR energy demand models were developed using population, export, import, and GDP data. Considering Turkey's socioeconomic situation, the energy demand was adjusted according to three different scenarios. Root Mean Square Error (RMSE), Mean Square Error (MSE), Mean Absolute Error (MAE), and R2 values were compared as error metrics to evaluate the performance results of the PSO, ANN, and SVR methods. When examining the error metric values, it can be said that ANN provides more successful results compared to the other methods. The correlation matrix was used to examine the relationship between energy consumption, which is the dependent output, and independent input parameters such as population, export, import, and GDP values that occurred between 1980 and 2019. A strong linear relationship between export and energy consumption was observed with a correlation value of 0.991 in the correlation matrix. Additionally, multiple regression equations were formed. Performance evaluation of prediction based on the four parameters (X1, X2, X3, X4) of import, export, GDP, and population was conducted using the F equation. The regression equation containing all four parameters had the highest R2 value of 0.995, indicating a comprehensive representation.