Stochastic Model Predictive Control of Combined Sewer Overflows in Sanitation Networks

Stokastisk Model Prædiktiv Kontrol af Overflod fra Kombinerede Kloakker i Sanitets Netværker

Knudsen, Casper Houtved ; Hodzic, Adis

4. term

Control and Automation, Master

2021

2021-06-03

126

Sanitation networks are essential infrastructure that transport wastewater and stormwater to treatment. Many places still use combined systems that carry both in the same pipe, which makes them prone to overflow during heavy rain. Real-time control can reduce overflows, and a common method is Model Predictive Control (MPC), which uses a model to look ahead and choose control actions. However, standard MPC struggles with uncertainty in inflow, especially from imperfect weather forecasts. To better handle uncertainty, we propose an extended framework called Chance-Constrained MPC (CC-MPC). We first formulate a nominal, multi-objective MPC tailored to the challenges of sewer operation, and then extend it to a stochastic MPC with chance constraints that limit the risk of violations (such as overflows) under uncertain forecasts. We compare the two controllers in a laboratory emulation of a network subsystem called the Two Tank Topology, which represents two connected tanks. Because flows are mainly driven by gravity, both controllers require a model that captures gravity pipe dynamics to predict overflows. We therefore develop a linear Diffusion Wave model based on discretized Saint-Venant partial differential equations. The model is validated using a data-driven parameter estimation framework. Identification is carried out both in a realistic network simulation and in a real-life experimental setup at the AAU Smart Water Lab.

Kloaksystemer er afgørende infrastruktur, der transporterer spildevand og regnvand til rensning. Mange steder bruges stadig fælleskloak, hvor begge typer vand løber i samme rør. Det gør systemet sårbart over for overløb ved kraftig nedbør. Realtidsstyring kan mindske overløb, og en udbredt metode er modelprædiktiv styring (MPC), som bruger en model til at forudsige og vælge styringshandlinger. Den klassiske MPC har dog svært ved at håndtere usikkerhed i indløbet, især fra usikre vejrudsigter. For bedre at håndtere usikkerhed foreslår vi en udvidet modelprædiktiv ramme kaldet chance-begrænset MPC (CC‑MPC). Vi opstiller først en grundlæggende MPC med flere mål, tilpasset udfordringerne i kloaksystemet, og udvider derefter rammen til en stokastisk MPC med sandsynlighedsbegrænsninger, som begrænser risikoen for f.eks. overløb under usikre prognoser. Vi sammenligner de to controllere i en laboratorieemulation af et delsystem kaldet To-tanks-topologien, som repræsenterer et netværk med to forbundne tanke. Da strømningen primært drives af tyngdekraften, kræver begge styringsmetoder en model, der kan beskrive rørdynamikken for at forudsige overløb. Derfor udvikler vi en lineær diffusionsbølgemodel baseret på diskretiserede Saint-Venant-ligninger. Modellen valideres via en databaseret ramme for parameterestimering. Identifikation gennemføres både i en realistisk netsimulering og i en virkelig forsøgsopstilling i AAU Smart Water Lab.

[This apstract has been rewritten with the help of AI based on the project's original abstract]

CC-MPC ; MPC ; SMPC ; sanitation networks ; sanitation ; sewer ; urban drainage networks ; sewer networks ; RTC ; chance costraints ; saint venant equations ; saint venant ; diffusion wave ; partial differential equation ; pde ; parameter estimation ; grey-box parameter identification ; data-driven parameter identification ; AAU Smart Water Lab ; Water Lab

Download PDF
View record in AAU Student Projects