Title: Predict-then-Optimise Strategies for Water Flow Control

Speaker: Vincent Barbosa Vaz

Summary:
We consider a common scenario for a pressure sewer system, where individual sites collect sewage in a local tank, and then pump it into the gravity fed sewage network. Standard control systems simply wait until the local tank reaches (near) capacity and begin pumping out. Unfortunately such simple control usually leads to peaks in sewage flow in morning and evening, corresponding to peak water usages in the properties. High peak flows require either large buffer pools or overflow systems, or larger capacity sewage treatment plants.

We investigate combining prediction and optimisation to better manage peak sewage flows. We use simple prediction methods to generate realistic possible future scenarios, and then develop optimisation models to generate pumping plans that try to smooth out flows into the network. The solutions of these models create a policy for pumping out specialized to individual properties, which overall is able to substantially reduce peak flows.

Biography:
Vincent is undertaking an OPTIMA Industry PhD Project with SE Water. His main research interests lie in machine learning and optimisation. His project area explores the interplay between these two fields and how they can be more tightly integrated since they have traditionally been performed independently. The overall aim is to investigate the deep integration of constrained optimisation and machine learning, and to provide more effective tools for society to tackle the challenging decision-making problems it is facing. The focus of his applied work is on adapting such a framework to the water sector to make more informed decisions.

He is supervised by OPTIMA CI Prof. Peter Stuckey, and Prof. James Bailey and Prof. Christopher Leckie at the University of Melbourne.

WED 18 OCTOBER 16:00-17:00 (AEDT, Melbourne Time)

MEETING ID: 873 1557 5255; PASSWORD: 778635