Learning enhanced situation perception for self-managed networks

Abstract

The networks in the future it is envisaged that they will be able to operate in an autonomous manner. In other words, it is consider that the networks will monitor their environment, analyze the environment stimuli, plan their operation and execute their plan. Moreover, the networks should be efficient and adaptable solutions so as to cover the diverse network requirements, ranging from typical human traffic, to ultra reliable or massive machine type traffic. This thesis aims at providing a scheme for situation aware networking, based on a hierarchical architecture, which enables the network elements to operate in a self managed way. We propose the introduction of two levels of hierarchy in the network management and control, the Network Element Controllers (NEC), and the Network Domain Controllers (NDC). The first ones have local network view and may proceed in handling of local problems, whereas the later have broader network view and may identify optimization opportunities or problems that are related to larger network compartments. Both NECs and NDCs are able to characterize their environment and identify their operational status, a functionality defined as situation perception. The development of the previously mentioned situation perception mechanisms, based on fuzzy reasoners, is the second major contribution of this thesis. The developed Situation schemes target QoS degradation events’ identification, Load events’ identification, and Cooperative power control. The third major contribution of this dissertation is the proposal of two learning schemes (a supervised one and an unsupervised one) for the enhancement of the situation perception mechanisms. The enhancement is related to the adaptation of the environment modeling of the fuzzy reasoners. This functionality enables the network elements to operate in new, unknown environments. The presentation of a generic reference problem, which is linked to the learning schemes, enables the application of these solutions in other problems, with similar formulations. Finally, the learning schemes are compared so as to provide directions on how these schemes may be used in other similar problems on the one hand, and what are the drawbacks and benefits of each scheme on the other.

Τα μελλοντικά δίκτυα θα έχουν τη δυνατότητα να λειτουργούν με αυτόνομο τρόπο. Τα συστήματα αυτά μπορούν να διαχειρίζονται ένα σύνολο από λειτουργίες, που περιλαμβάνουν να την των ελεγκήρων ασαφούς λογικής. Αναπτύχθηκαν δύο μηχανισμοί εκμάθησης που βασίζονται σε αλλαγή της μοντελοποίησης του περιβάλλοντος του μηχανισμού λήψης αποφάσεων. Ο πρώτος μηχανισμός εκμάθησης βασίζεται σε ένα σχήμα εποπτευμένης μάθησης, ενώ ο δεύτερος σε ένα σχήμα μη εποπτευμένης μάθησης. Οι δύο μηχανισμοί στηρίζονται σε μία βασική μοντελοποίηση του προβλήματος εκμάθησης (που επίσης προτείνεται στα πλαίσια αυτής της διατριβής) και αποσκοπεί στη γενίκευση των συγκεκριμένων μεθόδων εκμάθησης.