Analysis of frequent episodes in sequences of incidents collected in power distribution systems

Power distribution incidents collected by the electric utility through their customer service centers and incident management systems are analyzed systematically in this work. An incident is defined by the actions carried out by the utility to return the system to its normal operation when a failure occurrs. A sequence of incidents is a set of dated incidents recorded in a region or a network during a period of time, but only some subsets of them are of interest, either because they present similar features or because they are related in a way that allows them to be considered together (timing, periodicity, etc.). These subsets of significant incidents in a sequence are called episodes and are instances of possible patterns that define the behaviour of the system. The analysis is oriented to extract useful information from sequences of historical data to recognize relevant episodes that show a pattern of behaviour which could be useful to forecast future incidents or assist planning and operation. With this goal, the authors evaluate the use of a pattern discovery algorithm in sequences of incidents gathered in a network as part of the case study. Since the incidents can occur by random factors not all the frequent episodes that can be found are relevant. Then, it is proposed a methodology based on set theory to choose the relevant episodes which are especially useful to recognize outstanding episodes in sequences that have high randomness, which is the case of sequences of incidents in electrical networks.