Comparative analysis of metaheuristic optimization algorithms for natural gas demand forecast with meteorological parameters

Purpose: This study aims to present Turkey's monthly natural gas demand forecast model according to meteorological parameters with metaheuristic optimization algorithms.Theory and Methods: Meteorological data such as monthly average temperature, pressure, humidity, wind, precipitation were used as input parameters to estimate Turkey's natural gas demand. 2010-2017 data were used as training data and 2018-2020 data were used as test data. Four different metaheuristic algorithms (Artificial Bee Colony Algorithm (ABC), Charged System Search Algorithm (CSS), Crow Search Algorithm (CSA), and Harmony Search Algorithm (HSA)) and three different mathematical models (linear, exponential, quadratic) was used. Results were compared according to six different error metrics (AE, MAE, R2, MAPE, RMS, MARNE).Results: Natural gas demand estimation results from the models we propose based on meteorological parameters are as follows. In the linear model, the CSA algorithm gave the best result for the training dataset, while the ABC algorithm produced the best result for the test dataset. In the exponential model, while the algorithm that gave the best result in the training dataset was the HSA algorithm, the algorithm that gave the best result for the test dataset was the CSA algorithm. In the Quadratic model, while CSS produces a nominal error value for the training dataset, the CSA algorithm gives the best results in the test dataset. In general, the CSA algorithm gave the best result for the test data among the three mathematical models.Conclusion: As a result, four different metaheuristic algorithms have been successfully applied with three different models for Turkey's natural gas consumption estimation. Policymakers can use the results of the models presented in this study as a guide for future natural gas energy plans in Turkey. The proposed model can be extended by applying datasets from different countries and with different input parameters