Event Analysis on Power Communication Networks with Big Data for Maintenance Forms

High-Power Electric grid networks require extreme reliability in their associated telecommunication network to ensure protection and control throughout the power transportation. Whereas historical events are often available and carefully documented, few tools have been proposed to retrieve the information dealing with the heterogeneous data from maintenance forms and to support the network managers. In this work, we propose the use of Big Data and bootstrap resampling techniques in order to analyze the historical events on power communication networks based on the different data types that are more often used in maintenance forms, namely, numerical (such as duration), categorical (such as location), or text (such as anomaly description). Bootstrap difference tests are provided to create a similar framework for proportions, means, standard deviations, and bag-of-word distributions, and they provide us with an automatic significance level on the relevance of each field to measure the anomaly and failure severity in the communication network. Some few geographical regions and subregions were identified as prone to experiment more severe events, as well as some temporal patters corresponding to the week beginning and to the year ending. Bag-of-word analysis showed some few words being clearly present in more severe patterns, and their statistical distributions were sometimes multimodal, differently from the other features. Finally, duration analysis allowed us to quantify and to statistically describe the shorter duration of the event for the high-severity ones. These results offer a homogeneous framework for a single-feature based study in event analysis from maintenance forms in power communication networks, and they pave the way towards the principled use of more advanced multivariate techniques.