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

Forecasting natural gas demand is of great importance for the industrial sector, energy resources management, and especially for countries with high energy consumption. Estimating the natural gas consumption demand is important for making successful investments in the future in the decisions to be taken by decision-makers and managers. In this study, four different metaheuristic algorithms are used to estimate Turkey's natural gas demand. In our research, Differential Evolution Algorithm (DEA), Particle Swarm Search Optimization Algorithm (PSO), Gravity Search Algorithm (GSA), and Backtracking Search Optimization Algorithm (BSO) were compared according to their performance. The performances of the models were then evaluated with six different global error measurement approaches. Normalized meteorological data (average temperature, pressure, humidity, wind, and precipitation) were used as input parameters in the compared optimization models. Monthly (96 months) data between 2010-2017 were used as training data, and monthly (36 months) data between 2018-2020 were used as test data. Three mathematical models were used in the research: linear, exponential, and quadratic. According to the findings, the model that predicts real natural gas consumption data in all three models (linear, exponential, quadratic) among four different algorithms for training data is the second-order mathematical model of the DEA algorithm. For the test data, the model that predicts the most success in all three models among four different algorithms is the second-order mathematical model of the PSO algorithm.