Skip to Main content Skip to Navigation
Master Thesis

Apprentissage automatique d'automates temporisés à partir de séries temporelles

Abstract : The Time Automata (TA) is a formal method used to model dynamic systems such as ecosystems, electronic circuits, or security protocols. Numerous tools have been developed for the TAs, allowing to verify system properties expressed in mathematical logic with model checking or to generate test scenarios with controller synthesis. These models are currently built manually based on the experts' knowledge of the system. The work presented in this thesis concerns the use of data from sensors linked to a system for the modeling of the latter in the form of AT. This would make it possible to model systems which we do not know how they work. These sensor data often take the form of Time Series (TS). The problem is then the search for a method of automatic learning of TA from ST. With this aim in mind, the state of the art of TA machine learning algorithms is performed, as well as the state of the art of ST discretization methods because AT is based on discrete events. The algorithms are compared and a new algorithm, SYNTHESIS, is developed by taking the strengths of each one and trying to overcome their weaknesses. Different discretization methods of the state of the art are also compared on a synthetic dataset. The data sets from the discretization methods are used to learn TAs with the different algorithms including SYNTHESIS. The best results are obtain using a Hidden Markov Model or the k-Means to discretize the data, and SYNTHESIS to learn the TA. The main perspectives of this work are the formal verification of SYNTHESIS and the search for a validation tool for the learned TAs.
Document type :
Master Thesis
Complete list of metadatas
Contributor : Agrocampus Ouest <>
Submitted on : Thursday, October 15, 2020 - 9:02:30 AM
Last modification on : Sunday, January 10, 2021 - 3:08:50 AM


Files produced by the author(s)


Distributed under a Creative Commons Attribution - NonCommercial - NoDerivatives 4.0 International License


  • HAL Id : dumas-02967519, version 1



Lénaïg Cornanguer. Apprentissage automatique d'automates temporisés à partir de séries temporelles. Sciences du Vivant [q-bio]. 2020. ⟨dumas-02967519⟩



Record views


Files downloads