In order to understand energy consumption and ensure precise load prediction

it is essential to identify the variation of gas consumption in response to ambient temperature change outdoor. In this paper

the relationship is identified by using Empirical Mode Decomposition (EMD) and linear regression analysis together with outlier detection. EMD is a data processing tool that can divide original data into several Intrinsic Mode Functions (IMFs) with a lower frequency residue. By applying the data mining technique-Mahalanobis distance measurement

some outliers from real-time gas consumption and temperature data points are detected

which are excluded from the data sets to ensure accuracy. Correlation coefficients between the gas load and ambient temperature are calculated and denoted as an important index to quantify their relationship through regression analysis. By comparing such indices on realtime data and EMD processed data

the weather-sensitive part of gas demand is identified. The methods are implemented on a local energy system and the results reveal that the outcome after EMD presents a higher level of correlation between the gas load and ambient temperature

compared to the results from directly using the real-time gas load and temperature data.