The information dynamics of cascading failures in energy networks

Disclaimer: shameless self-promotion follows.

So our submission:

Joseph T. Lizier, Mikhail Prokopenko, David J. Cornforth, "The information dynamics of cascading failures in energy networks"

to ECCS 2009 was accepted, and I'll be presenting it in the Policy, Planning and Infrastructure track currently on the Friday morning of the conference.

The abstract is as follows:

Small failures in electrical energy networks can lead to cascading failures that cause large and sustained power blackouts. These can disrupt important services and cost millions of dollars. It is important to understand these events so that they may be avoided. We use an existing model for cascading failures to study the information dynamics in these events, where the network is collectively computing a new stable distribution of flows. In particular, information transfer and storage across the network are shown to exhibit sensitivity to reduced network capacity earlier than network efficiency does, and so could be a useful indicator of critical loading. We also show that the local information dynamics at each node reveals interesting relationships between local topological features and computational traits. Finally, we demonstrate a peak in local information transfer in time coinciding with the height of the cascade's spread.

In a nutshell, this paper describes an application of our framework for the information dynamics of distributed computation to the phenomena of cascading failures on networks. The focus is on energy networks, though the results are applicable to other types of networks, e.g. transport.

Information dynamics may at first not seem applicably to cascading failures, but there are a few good reasons for the application here. First, cascading failures are akin to damage spreading phenomena, and both are often cited as mechanisms of information transfer in networks: it is useful to explore this quantitatively. Further, when a cascading failure occurs, the network is actually computing a new stable state (or attractor), so quantifying the information dynamics is a direct study of this computation. To underline all that, I really like this quote from Melanie Mitchell's new book:

The phenomena of cascading failures emphasizes the need to understand information spreading and how it is affected by network structure.

Primarily, the results show that we get maximisations of information transfer and storage in the network near the critical phase, aligning with our findings in Random Boolean Networks (RBNs) in a paper at ALifeXI last year. We also find some interesting relationships between topological properties of the individual nodes and their own local information dynamics.

From here, I'll be combining this work with that on RBNs in my PhD thesis, and probably seeking to make a journal submission from their combination.