Quantifying the Benefits of using Machine Learning for the Smart Design and Control of Stormwater Storages

Abstract

Urban stormwater systems are under increasing stress due to densification and climate change, often necessitating expensive pipe upgrades. In addition to the high cost of these upgrades, they are often not feasible due to practical constraints (e.g. proximity of other services, traffic disruption, etc.). An alternative is to reduce peak system flows via storage. Traditionally, such storage is placed at catchment outlets, requiring significant volumes and space to be effective. The benefits of a given storage volume can be increased significantly by distributing a series of smaller storages throughout catchments at strategic locations, as these not only reduce peak flows locally, but can reduce the coincidence of hydrographs from contributing sub-catchments. The ‘Smart Design’ approach to distributed stormwater storage outlined in this paper uses machine learning multi-objective optimisation to identify layouts (number of storages, their locations and sizes, and outlet size) that minimise system peak flows at critical locations under a range of design conditions. This enables Pareto-optimal layouts to be identified from the large number of potential layouts. ‘Smart Control’ is where controls are added to the outlets of these storages and by operating them in real-time during storm events system peak flows can be further reduced. In addition, such smart controls provide opportunities for stormwater re-use for urban greening ‘Smart design’ and ‘Smart control’ (as described above) is implemented on a series of case studies, to quantify the practical benefits of this approach. The case studies illustrate how ‘smart design’ can save space (45%-60%), reduce costs (30%-60%) and is more feasible than traditional approaches by using smaller storages that fit into existing space. The case studies also illustrate how ‘smart control’ can further reduce peak flows by an additional 10%-20% for the minor design event (10% AEP). One case study also demonstrated the potential for smart control to adapt to more extreme events by providing peak flow reductions for 1% AEP events. These outcomes demonstrate that smart design and control of distributed storages provides an opportunity to reduce costs, save space and increase the feasibility of stormwater systems to address the future challenges of increased peak flows due to urban densification and climate change. We welcome opportunities for further case studies to demonstrate if these findings hold for a wider range of catchments and climates.