Short-term forecasting of natural gas consumption by determining the statistical distribution of consumption data

Abstract

The development of gas smart meters has enabled the collection of data on daily natural gas consumption which can be used to develop and improve methods and models for natural gas consumption forecasting. This paper presents the development of a model for the short-term forecasting of total natural gas consumption, which is applicable in different distribution areas where smart meters are installed in large numbers. The advantages of this model are the use of only two input parameters (daily natural gas consumption and average daily temperature), forecasting the total consumption in the determined area by analyzing the consumption data of less than 10% of the total consumers as well as robustness to consumer types. Daily natural gas consumption data collected from the more than 3300 gas smart meters over a period of six months was used for the determination of correlations between lognormal distribution variables and temperature. The defined correlations between distribution variables and temperature were used for upscaling consumption to a specific number of final consumers, i.e., to obtain the total consumption of natural gas in the observed area. Best results were achieved using the “two-day rolling average temperature” in the consumption scenario up to 250 m3 per day (MAPE was 7.26%). When compared to using “average temperature” as an input parameter, “two-day rolling average temperature” and “shaving peaks temperature” produced better results due to the mitigated impact of sudden temperature changes that significantly affected the simulated consumption in the model while the actual consumption is a little more inert. Also, consumption scenarios up to 250 m3 can be considered the most representative for forecasting total natural gas consumption since it achieved the best results.