טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentFraid Oded
SubjectDynamics of Cascading Failure from an Info-Gap Perspective
DepartmentDepartment of Mechanical Engineering
Supervisor Professor Yakov Ben-Haim
Full Thesis textFull thesis text - English Version


Abstract

Networks are a crucial part of our day to day life. We encounter them in all realms of life and when they crash the results are often catastrophic. This thesis investigates the benefits to be gained by using the info-gap approach in order to examine cascading failure in networks. 

               A new way to model a network is introduced; one that is driven by the actual supply and demand of the system and takes into account possible uncertainty in the system. This model consists of nodes and arcs which represent a general network. In that network each node communicates to its closest neighbors the gap between its demand and its consumption. That gap is the need it “feels” and “wants” to fulfill. The proposed dynamic model is an iterative one that changes the current flow on the arcs, based on the gap between the demand and the consumption. The model tries to find a working point in which the consumption of the nodes meets their demand. In the proposed model, uncertainty is introduced in the real capacity of the arcs. Other types of uncertainty could be considered like uncertainty in the supply or the demand of the nodes or in the exact consumption of the nodes.

The first benefit of using this new proposed modeling scheme is that cascading failure occurs only when uncertainty is introduced into the system. It is a common design requirement for a system to be stable under the conditions for which it was designed. Failure may only occur when the unexpected happens. However, the unexpected occurs quite frequently. This thesis demonstrated the evaluation of the vulnerability to failure as a result of unexpected occurrences, while assuming only very limited prior information about these surprises.

 The second benefit of using the new proposed modeling is the ability to compare strategies. By comparing strategies it is shown that even in the simple cases that were examined a reversal of preference between strategies may occur. This preference reversal should be particularly interesting to the designers of the network who can now choose a strategy that allows more robustness with the same resources.