Urban Drainage Systems PhD Candidate: Mahmood Mahmoodian Daily - - PowerPoint PPT Presentation

urban drainage systems
SMART_READER_LITE
LIVE PREVIEW

Urban Drainage Systems PhD Candidate: Mahmood Mahmoodian Daily - - PowerPoint PPT Presentation

Surrogate Modelling for Urban Drainage Systems PhD Candidate: Mahmood Mahmoodian Daily supervisor: Ulrich Leopold WHAT IS SURROGATE MODELLING? An analogy Detailed rabbit simulator (model) Real rabbit Surrogate model 1 Surrogate model 2


slide-1
SLIDE 1

PhD Candidate: Mahmood Mahmoodian Daily supervisor: Ulrich Leopold

Surrogate Modelling for Urban Drainage Systems

slide-2
SLIDE 2

An analogy

WHAT IS SURROGATE MODELLING?

2

Real rabbit

Detailed rabbit simulator (model)

Surrogate model 2 Surrogate model 3 Surrogate model 1

slide-3
SLIDE 3

3

MY CHALLENGE

Urban Drainage System

Computationally expensive for:

  • Real-time control (RTC)
  • Calibration
  • Optimisation
  • Uncertainty propagation

Zoom in Modelling

WWTP

Storage Tank 1 Storage Tank 2

slide-4
SLIDE 4

To obtain fast simulators

POSSIBLE SOLUTIONS:

4

Develop a surrogate of already existing simulators (e.g. InfoWorks,…)

  • 1. Data-driven approaches
  • 2. Projection-based approaches
  • 3. Hierarchical or multi-fidelity approaches
  • 4. Hybrid approaches
slide-5
SLIDE 5

METHOD 2: A DATA-DRIVEN APPROACH

5

GAUSSIAN PROCESS EMULATOR (GPE) Main advantages:

  • Non-intrusive and generic method
  • Uncertainty quantification

Main disadvantages:

  • Limited number of parameters
  • Short simulations

Validation Training data:

Olson & Chang (2013)

Y: model output of interest µβ : a mean function which is considered linear in time ξy : a vector of covariance matrix parameters. Validation

slide-6
SLIDE 6

METHOD 2: A HYBRID METHOD (MODEL-DRIVEN + DATA-DRIVEN)

6

Model-driven Data-driven

Storage Tank Volume Change:

Main advantages:

  • Super fast! (1400 times faster)
  • Useful for longer simulations

Main disadvantages:

  • Partly-intrusive and case-specific
  • Uncertainty quantification?

Validation

mass balance

slide-7
SLIDE 7

7

TAKE AWAY MESSAGE

There is NO generic remedy of surrogate modelling for all applications Selection criteria:

  • 1. Reduce the runtime significantly
  • 2. Give a quantification of the uncertainty
  • 3. Be easily applicable in practice

Example: real-time control (RTC) Data-driven GPE approach!

slide-8
SLIDE 8

8

THANK YOU FOR YOUR ATTENTION!

Keep It as Simple as Possible! (Law of parsimony, Ockham's Razor) Email: mahmood.mahmoodian@list.lu

slide-9
SLIDE 9

9

PARTNERS AND ACKNOWLEDGEMENTS

This project has received funding from the European Union’s Seventh Framework Programme for research, technological development and demonstration under grant agreement no 607000.

www.quics.eu