Machine Learning for Short-Term Water Demand Predictions

Urban water infrastructure plays a central role in our societies. Now it has become ever more important in the reliable provision of safe drinking water and effective drainage and sewerage services due to increasing pressures from many sources: growing population, demographic change, stricter water quality requirements and changing climate. In the meantime, water system planning and management has to address many different, often conflicting needs from multiple stakeholders, such as costs, environmental impacts, water quality, reliability, risk, and resilience to various chronical and acute threats. The focus of the infrastructure analysis and optimization work over the past 20 years has been on multi-objective problems, considering two to three objectives in their formulation. However, it is known that for many-objective optimization, where there are typically more than three objectives, the standard Pareto-optimal algorithms may lose their effectiveness. Thus, there is a need for a new paradigm based on many-objective analysis and visual analytics in order to provide informed decision making for complex water management problems.

This project aims to develop a new many-objective analysis approach that can provide useful and practical decision analytics for water infrastructure systems. It will tackle the challenges in improving the search efficiency and quality of many-objective evolutionary algorithms and visualizing high-dimensional Pareto fronts to support decision making. New approaches will be investigated to incorporate uncertainty quantification into the decision analytics tools in an effective manner. To test the new approach and tools, benchmark and real world case studies from water supply, urban drainage and/or wastewater systems will be studied to explore the advantages and disadvantages of multiple design and management options.

This project will produce a new approach and a suite of decision analytics tools that could be used by researchers, practical engineers and decision makers to solve complex water infrastructure management problems. It will build on the previous research by the supervisory team, in particularly the following award winning studies: 1) water distribution system design (Fu et al. 2013), which won the ASCE 'Quentin Martin Best Practice Oriented Paper' award for 2014; 2) Parallel Evolutionary Optimization (Guidolin et al 2012), which won double prizes (the international BWN Competition and the best paper award presented by Prof Dragan Savic ); 3) multi-objective control of wastewater treatment plants (Sweetapple et al. 2014), the first author (supervised by Dr Fu) won the Scopus Young Researcher Awards for 2015.

Fu G, Kapelan Z, Kasprzyk J, Reed P. (2013). Journal of Water Resources Planning and Management, 139(6), 624-633.
Guidolin M, Fu G, Reed P. Savic D. Water Distribution Systems Analysis 2012, Australia.
Sweetapple C, Fu G, Butler D. (2014). Water Res, 55: 52-62, DOI:10.1016/j.watres.2014.02.018.