Forecasting Residential Consumption of Natural Gas Using Genetic Algorithms

Abstract

This paper presents an application of genetic algorithms to forecast short-term demand of natural gas in residences. Residential demand is assumed to be a function of time, heating degree-day value, and consumer price index. A genetic algorithm is designed to estimate parameters of a multiple nonlinear regression model which mathematically represents the relationship between natural gas consumption and influential variables. Genetic algorithms have recently received attention as robust stochastic search algorithms to solve various forecasting problems since they have several significant advantages over conventional methods. Without requiring assumptions need to be made about the underlying function or model, genetic algorithms can attain proper solutions by scanning solution space from many different starting point. To show the applicability and superiority of the described approach, it is considered the monthly data of the residential sector which consumes 23% of imported natural gas in Turkey. The results have revealed that genetic algorithms can be used as an alternative solution approach to forecast the future demand of natural gas.