Integration of weather forecast and artificial intelligence for a short-term city-scale natural gas consumption prediction

Abstract

Ruled by the directives and regulations adopted by the European Commission, the European natural gas market is obliged to enable free competition. This process has just started in the Serbian market, and the national natural gas provider (Srbijagas) has started its restructuring. To attain the objective of natural gas market liberalization, it is essential to accurately plan capacities and make precise predictions of future peak consumption at all levels of trade. In this research, an hourly city-scale Adaptive Neuro-Fuzzy Inference System algorithm with the Gaussian membership function is developed for both heating and non-heating regimes. In order to upgrade previous studies in this field, besides standard gas consumption and weather variables, hourly weather forecast and lower calorific value were added to the model. The model output is predicted hour-ahead natural gas consumption. To measure the predicting performance of the two models proposed in this research, the Coefficient of Determination and Mean Absolute Percentage Error were used. For the considered hourly consumption forecast of heating and non-heating models, the Coefficients of Determination were 0.99, while the overall (training, testing and checking) Mean Absolute Percentage Error values were 3.0% and 3.4% respectively. Based on these performance metrics, it can be concluded that for both models, the proposed algorithm resulted in a very satisfactory forecast. An adequate prediction of future natural gas consumption would support all market stakeholders, leading to the development a cleaner and sustainable energy system.