Chapter 13: Empirical observations and statistical analysis of gas demand data

Abstract

In this chapter we describe an approach for the statistical analysis of gas demand data. The objective is to model temperature dependent univariate and multivariate distributions allowing for later evaluation of network constellations with respect to the probability of demand satisfaction. In the first part, methodologies of descriptive data analysis (statistical tests, visual tools) are presented and dominating distribution types identified. Then, an automated procedure for assigning a particular distribution to the measurement data of some exit point is proposed. The univariate analysis subsequently serves as the basis for establishing an approximate multivariate model characterizing the statistics of the network as a whole. Special attention is paid to the statistical model in the low temperature range. The goal of our data analysis consists in evaluating historical data on gas demand at exits of some gas transportation network. The results will be used to extract statistical information, which may be exploited later for modeling the gas flow in the network under similar temperature conditions. More precisely, the aim is to generate a number of scenarios of possible exit loads, which will be complemented in several subsequent steps to complete a nomination (see Chapter 14). Such scenarios are needed for validating the gas network and for calculating and maximizing its technical capacities. The analysis will be based on historical measurement data for gas consumption, which is typically available during some time period, and on daily mean temperature data provided by a local weather service. Due to a high temperature-dependent proportion of heating gas, the gas demand is subject to seasonal fluctuations. During the warmer season the gas consumption decreases: hot water supply for households and process gas consumption are the only basic constituents. The method for analyzing the data should be applicable to all exits, no matter what their distribution characteristics are, and should allow for multivariate modeling to take into account statistical dependencies of different exits of the network. Therefore, the use of local temperatures as in day-ahead prediction of gas demands is less appropriate. Rather, we introduce a reference temperature which is given as a weighted sum of several local