Personal tools
Laboratoire d'InfoRmatique en Image et Systèmes d'information

Skip to content. | Skip to navigation

Laboratoire d'InfoRmatique en Image et Systèmes d'information
UMR 5205 CNRS / INSA Lyon / Université Claude Bernard Lyon 1 / Université Lumière Lyon 2 / École Centrale de Lyon
You are here: Home > membres

Lucas Foulon

photo

PhD student

Team Data Mining and Machine Learning
 
Institution (SNCF)
Location Blaise Pascal (INSA)
 
E-mail lucas.foulon at liris.cnrs.fr
URL http://liris.cnrs.fr/lucas.foulon
Contact details Publications Thesis
 
Subject Outlier detection in real-time flow ground-aboard on the SNCF
Abstract SNCF produces and operates in its information system a large amount of heterogeneous data processed in real time.
Some of them originates from "ground" information system, like for example real time passenger information (upcoming departures, disturbances, etc.), and others come from the board, produced by so-called "communicating trains" (such as geo-location data, tele-maintenance, train state, etc.).
All these data flows are collected in real time, aggregated, standardized and distributed by mediation platforms or service buses. They are emitted by different types of embedded devices, circulate between and are processed by different platforms, before arriving to client applications.
These flows require an end-to-end monitoring, to observe their behavior and many variations in data traffic.
These variations are caused by the dynamics of the railway system (eg. traffic disruption), or related to the processing and communication infrastructure used to process and transmit the data.
This thesis is focused on this very last point, namely detecting processing and communication infrastructure technical anomalies, rather than railway system disturbance.
Indeed, when such an anomaly occurs, the delay required to detect it can sometimes reach several days, or even weeks, as well as the time necessary to identify the cause in order to resolve it.
This thesis is a continuation of a Master 2 stage in which a first flow modeling and an anomaly detection algorithm have been set up for the train geolocation data flow.
The objective of this thesis is to have a complete dynamic model of the real-time data flow "normality", capable of adapting to regim shift, thanks to the theoretical knowledge on the network and to its normal activity, while modeling the correlation relations and causality between the various indicators, with the aim of detecting anomalies (viewed as a difference or distance form the normality), to better locate and prevent them.
Advisor Serge Fenet
Advisor Christophe Rigotti

Last update : 2017-10-20 16:31:37