Decision-making under Uncertainty for COmplex systems (DUCO) team

Our vision: Bringing uncertainty-awareness and systems-thinking into everyday decisions.

We study uncertainty quantification (UQ) and decision-making in complex systems. Our vision is to make UQ and decision tools part of everyday decisions by overcoming current barriers, including high computational costs, memory demands, expertise requirements, and validation challenges. Realising this vision will not only reduce inefficiencies caused by unmanaged system complexity but also unlock the potential of complexity as a source of redundancy and resilience.
We aim to develop general tools applicable across diverse systems. Our interests include, but are not limited to, ageing structures and assets, transport networks, energy grids, and process plants.

Our research

Matrix-based Bayesian network (MBN)

MBN and the Python toolkit MBNPy are a tool to enable Bayesian network (BN) to handle large-scale systems. BN is a useful tool to visualise complex probabilistic dependence between various types of variables (see the figure on the left-hand side).

MBN solves the memory issue arising from the converging structure between components and the system. This is done by its encoding scheme of the high-dimensional probability distribution $P(S | X_1,\cdots,X_N)$

With MBN, one can perform probabilistic analysis of complex AND large-scale system problems. See for example, the figure below.
Possible analyses include reliability assessment, component importance measure, Value of Information, Bayesian inference of inspection results, and uncertainty-based optimisation.

Buffered optimisation and reliability method (BORM)

Reliabiltiy-based optimisation (RBO) aims to find the most economical solution that satisfies the reliability requirement; see the figure below.

As a tool for RBO, BORM greatly improves computational efficiency, especially in data-driven settings (Byun and Royset 2022) and given a general, large-scale system (Byun et al. 2023). The large power system in the figure below has been solved by BORM within 1.5 hours with a personal laptop.

If interested, check out the Matlab codes at GitHub. We proudly say that it works quite well.