Student research opportunities
Achieving diagnosability by AI planning, with application to the Smart Grid
Project Code: CECS_134
This project is available at the following levels:
Honours, Summer Scholar, Masters, PhD
Keywords:
AI, AI planning, Diagnosis, Diagnosability, Smart Grid
Supervisors:
Dr Adi BOTEADr Alban Grastien
Outline:
This project takes place within the hot topic of Smart Grids for which
many countries around the world (including Australia) are investing
billions of dollars.
Planning is the problem of generating a sequence of actions that, if
performed by an (automated) agent, will lead the system to a state
with specific properties (the /goal/). Most academic work gives this
goal a very simple definition (e. g. a simple combination of atomic
propositions). The goal can however be more complex and may consist
in leading the system to a state where a temporal specification is
satisfied. For instance, an electricity distribution network is often
tolerant to failures but can collapse after a certain number
(typically as few as two) of failures take place; it is important to
consider this type of fault-tolerance specifications when a
reconfiguration plan is generated.
This project focusses on a different property that we want to enforce.
An important aspect of monitoring complex systems is the diagnosis,
i.e. the detection and identification of failures in the network.
Being able to accurately and quickly determine the occurrence of
faults in a network allows for rapid intervention leading to 1) faster
restoration and 2) lower vulnerability to multiple faults. The
capacity to identify faults is called diagnosability.
Goals of this project
The aim of the project is to combine a planning technique developed
by Adi Botea and André A. Ciré [1] where the goal is defined as a
specification in LTL, with a diagnosability technique developed by
Alban Grastien [2] where the diagnosability testing is specified in
terms of an LTL formula.
Background Literature
[1] André A. Ciré and Adi Botea. 2008. Learning in Planning with
Temporally Extended Goals and Uncontrollable Events. In Proceedings of
the European Conference on Artificial Intelligence ECAI-08. Patras,
Greece.
[2] Alban Grastien. Symbolic Testing of Diagnosability. In
International Workshop on Principles of Diagnosis (DX-09), pages
131-138, 2009.
