Learning partial-value variable relations for system modeling

Abstract

An important part of system modeling involves establishing relations of system variables. Data collected from a system reflects relations of system variables and thus allows us to learn variable relations from system data. Existing machine learning and data mining techniques focus on learning variable relations that hold for all values of variables. However, different variable relations may exist for different ranges of variable values, or a variable relation holds only for certain ranges of variable values but not for full ranges of variable values. This paper presents the use of a new algorithm, called Partial-Value Association Discovery (PVAD), to learn partial-value variable relations for energy consumption system modeling and engineering retention.