Smart Real-time Control of Water Systems
Henrik Madsen, Peter Steen Mikkelsen, Lasse Engbo Christiansen, Anne Katrine Falk, Morten Borup, Rune Juhl, Nadia
Schou Vorndran Lund, Rasmus Halvgaard, Nina Donna Sto. Domingo, Lisbeth Birch Pedersen, Stephen J. Flood & Lene
Bassøe
Urban Drainage Group Autumn Conference and Exhibition 2016
Blackpool, November 9th – 11th
Dr. Lisbeth Birch Pedersen
Product Owner
MIKE Powered by DHI
Outline
• MPC and surrogate models in real-time control of urban water
systems
• Aarhus – full scale test and implementation
• Status and what’s next
© DHI #2
Who am I?
© DHI
Tax payer
Project consultant
End user
#3
Motivation for MPC and surrogate models
in real-time control of urban water systems
© DHI #4
Integrated Urban Water Management
© DHI #5
Different Solutions for Different Maturity Levels and Different
Challenges
Optimised
MPC
Forecast model
On-line model
On-line Data
© DHI #6
© DHI
Local control
System-wide Model Predictive Control (MPC)
#7
Why think of a new optimisation approach?
© DHI
Current optimisation system
• Uses simplified optimisation
model (few decision variables)
• Includes execution of many
hundreds of MIKE 11
simulations
• Takes hours on a 16-core
server
New optimisation system
• Uses detailed optimisation
model (thousands of decision
variables)
• Model dynamics described by
a simplified (surrogate) model
• Takes few minutes on today’s
laptop
#8
What is a surrogate model?
• Derived from the HiFi model (MIKE model)
• Sufficiently accurate for modelling the most important characteristics relevant for the
problem at hand
• Computationally fast
© DHI
Automated
conceptualisation
#9
MPC-surrogate modelling framework
© DHI
Modular framework for building surrogate models
Qin
Qout
Qlat
Qin
Qout
Qout
Qin
#10
MPC-surrogate modelling framework
© DHI
MIKE
model
Surrogate
model
1
2
Formulation of optimisation model
• Constraints
• Operation targets
• Objective function
3
Behind the scenes
• Automatic setup of MPC optimisation model
• Efficient optimisation solver
#11
Benefits
• New technology enables solving large system-wide optimisation and control problems
in real-time
− Problems that we cannot solve today
• Significant value propositions within several business areas
− Creating value to real-time forecasting and operations solutions and services
• The generic basis of the technology allows to efficiently develop new solutions within
new areas.
© DHI
More foodReduced flooding Environmental
protection
More hydropower More money
#14
From vision to operation
Real-time control of urban drainage system,
Aarhus, Denmark
© DHI #15
© DHI
Drivers
…Integrating
water into the
urban space
… Recreational
use of water
…New housing
area on the
harbor front
…Rapid city
development
©Politiken©Aarhus.dk
#16
DHI & Aarhus Water - Leading the way
© DHI
Samstyrring I
• Analysis and design
• 2006-2007
• More Info
Samstyrring II
• Implementation of
infrastructure
• 2007-2012
• More Info
Samstyrring III
• Integrated modelbased
control and warning
• PREPARED
• 2009-2013
• More Info
Klimaspring
• MPC and surrogate
modelling for real-time
control
• REAL-DANIA
• 2015-2017
• More Info
Water Smart Cities
• Climate change adaptation
through intelligent
software technologies
• MPC in real-time control
• 2016-2020
• More info
#17
Prerequisites for Smart Real-Time Control
• Remote controllable elements
• SCADA system
• Data integration in real-time across “silos”
• Common framework for automatic execution of data and models
© DHI #18
© DHI
Physical system
HiFi model (MIKE model)
Optimiser
Allows formulation of simplified
optimisation
Physical system
HiFi model (MIKE model)
Optimiser
Allows formulation of very large
optimisation problem
Surrogate model (Linear)
MPC optimisation framework
100.000’s of decision variables10’s of decision variables
#19
© DHI
HiFi
system state
Levels
Flows
Gate positions
Optimised
control set-points
Gates, pumps, valves
MIKE URBAN
RR
CATCHMENT
RUNOFF
MIKE URBAN
HD
Rainfall
(incl. forecast)
HiFi Models
Optimizer
Surrogate
model
RR
Model
CATCHMENT
RUNOFF
Surrogate
Models
Surrogate model
Initial state data
#20
© DHI
Rainfall forecast
ensemble
Radar data processing:
• Conversion to
catchment rainfall
Surrogate rainfall-
runoff model
Catchment runoff
time series
MPC control
model
Optimised control
Observed radar and
rain gauge data
Radar data processing:
• dBZ adjustment
• Bias adjustment
Ensemble radar
nowcast model
MIKE URBAN
model
Operation data
Water level/flow
measurements
Data assimilation
system
Radar data processing:
• Conversion to
catchment rainfall
MIKE OPERATIONS workflow
#21
© DHI
Layer 3
PLC/SCADA
Sensors/Actuators
WISYS
Short-term ensemble
rainfall forecast
Global Model
Predictive Control Optimised control
WWTP max.
hydraulic load
Data validation and
filtration
Software sensors:
Flow, Elevations,
tank filling, etc.
PID (flows) at each
storage tank
PID (elevations) at
each storage tank
PID output: 0-100%
distributed to set-
points for pumps,
weirs/gates at each
storage tank
Layer 2
Layer 1
Levels, flows and
weir/gate positions Set-points
Real-Time Control System
Set-points
#22
© DHI
Control action
Output variable(s)
Uncontrolled boundaries
Radar-rainfall
Rainfall-runoff
model
Control
model
#23
© DHI #24
Trøjborg overflow model validation
© DHI #25
Basin water
level Overflow
discharge
Outflow set
point
Status and what to come
© DHI #26
© DHI
Consolidated in…
#27
Currently ongoing at DHI
• Working MPC - surrogate modelling framework within MIKE WORKBENCH
• Demonstration of framework for reservoir flood control and optimisation of large-scale
irrigation system
• First tests of control of urban storm and wastewater systems
© DHI
2. MPC
1. RR
3. HD-HiFi
0. Rainfall-
Forecast
#28
Water Smart Cities project 2016-2020
• MPC-surrogate modelling framework
− Extension of surrogate model class
− Extension of methods for handling forecast uncertainty (incl. short to medium range weather
forecasts)
− Adaptive control automatically shifting between different control strategies depending on
system state and rainfall forecast
− Combined control of drainage system and WWTP
• Forecast models
− Probabilistic forecasting
− Use of surrogate models
− Data assimilation
• Automatic surrogate model builder
− MIKE URBAN -> surrogate model suitable for different purposes (e.g. forecast, control)
• MIKE OPERATIONS implementation
© DHI #29
Thank you
Dr. Lisbeth Birch Pedersen
lpe@dhigroup.com
© DHI #30

Smart Real-time Control of Water Systems

  • 1.
    Smart Real-time Controlof Water Systems Henrik Madsen, Peter Steen Mikkelsen, Lasse Engbo Christiansen, Anne Katrine Falk, Morten Borup, Rune Juhl, Nadia Schou Vorndran Lund, Rasmus Halvgaard, Nina Donna Sto. Domingo, Lisbeth Birch Pedersen, Stephen J. Flood & Lene Bassøe Urban Drainage Group Autumn Conference and Exhibition 2016 Blackpool, November 9th – 11th Dr. Lisbeth Birch Pedersen Product Owner MIKE Powered by DHI
  • 2.
    Outline • MPC andsurrogate models in real-time control of urban water systems • Aarhus – full scale test and implementation • Status and what’s next © DHI #2
  • 3.
    Who am I? ©DHI Tax payer Project consultant End user #3
  • 4.
    Motivation for MPCand surrogate models in real-time control of urban water systems © DHI #4
  • 5.
    Integrated Urban WaterManagement © DHI #5
  • 6.
    Different Solutions forDifferent Maturity Levels and Different Challenges Optimised MPC Forecast model On-line model On-line Data © DHI #6
  • 7.
    © DHI Local control System-wideModel Predictive Control (MPC) #7
  • 8.
    Why think ofa new optimisation approach? © DHI Current optimisation system • Uses simplified optimisation model (few decision variables) • Includes execution of many hundreds of MIKE 11 simulations • Takes hours on a 16-core server New optimisation system • Uses detailed optimisation model (thousands of decision variables) • Model dynamics described by a simplified (surrogate) model • Takes few minutes on today’s laptop #8
  • 9.
    What is asurrogate model? • Derived from the HiFi model (MIKE model) • Sufficiently accurate for modelling the most important characteristics relevant for the problem at hand • Computationally fast © DHI Automated conceptualisation #9
  • 10.
    MPC-surrogate modelling framework ©DHI Modular framework for building surrogate models Qin Qout Qlat Qin Qout Qout Qin #10
  • 11.
    MPC-surrogate modelling framework ©DHI MIKE model Surrogate model 1 2 Formulation of optimisation model • Constraints • Operation targets • Objective function 3 Behind the scenes • Automatic setup of MPC optimisation model • Efficient optimisation solver #11
  • 12.
    Benefits • New technologyenables solving large system-wide optimisation and control problems in real-time − Problems that we cannot solve today • Significant value propositions within several business areas − Creating value to real-time forecasting and operations solutions and services • The generic basis of the technology allows to efficiently develop new solutions within new areas. © DHI More foodReduced flooding Environmental protection More hydropower More money #14
  • 13.
    From vision tooperation Real-time control of urban drainage system, Aarhus, Denmark © DHI #15
  • 14.
    © DHI Drivers …Integrating water intothe urban space … Recreational use of water …New housing area on the harbor front …Rapid city development ©Politiken©Aarhus.dk #16
  • 15.
    DHI & AarhusWater - Leading the way © DHI Samstyrring I • Analysis and design • 2006-2007 • More Info Samstyrring II • Implementation of infrastructure • 2007-2012 • More Info Samstyrring III • Integrated modelbased control and warning • PREPARED • 2009-2013 • More Info Klimaspring • MPC and surrogate modelling for real-time control • REAL-DANIA • 2015-2017 • More Info Water Smart Cities • Climate change adaptation through intelligent software technologies • MPC in real-time control • 2016-2020 • More info #17
  • 16.
    Prerequisites for SmartReal-Time Control • Remote controllable elements • SCADA system • Data integration in real-time across “silos” • Common framework for automatic execution of data and models © DHI #18
  • 17.
    © DHI Physical system HiFimodel (MIKE model) Optimiser Allows formulation of simplified optimisation Physical system HiFi model (MIKE model) Optimiser Allows formulation of very large optimisation problem Surrogate model (Linear) MPC optimisation framework 100.000’s of decision variables10’s of decision variables #19
  • 18.
    © DHI HiFi system state Levels Flows Gatepositions Optimised control set-points Gates, pumps, valves MIKE URBAN RR CATCHMENT RUNOFF MIKE URBAN HD Rainfall (incl. forecast) HiFi Models Optimizer Surrogate model RR Model CATCHMENT RUNOFF Surrogate Models Surrogate model Initial state data #20
  • 19.
    © DHI Rainfall forecast ensemble Radardata processing: • Conversion to catchment rainfall Surrogate rainfall- runoff model Catchment runoff time series MPC control model Optimised control Observed radar and rain gauge data Radar data processing: • dBZ adjustment • Bias adjustment Ensemble radar nowcast model MIKE URBAN model Operation data Water level/flow measurements Data assimilation system Radar data processing: • Conversion to catchment rainfall MIKE OPERATIONS workflow #21
  • 20.
    © DHI Layer 3 PLC/SCADA Sensors/Actuators WISYS Short-termensemble rainfall forecast Global Model Predictive Control Optimised control WWTP max. hydraulic load Data validation and filtration Software sensors: Flow, Elevations, tank filling, etc. PID (flows) at each storage tank PID (elevations) at each storage tank PID output: 0-100% distributed to set- points for pumps, weirs/gates at each storage tank Layer 2 Layer 1 Levels, flows and weir/gate positions Set-points Real-Time Control System Set-points #22
  • 21.
    © DHI Control action Outputvariable(s) Uncontrolled boundaries Radar-rainfall Rainfall-runoff model Control model #23
  • 22.
  • 23.
    Trøjborg overflow modelvalidation © DHI #25 Basin water level Overflow discharge Outflow set point
  • 24.
    Status and whatto come © DHI #26
  • 25.
  • 26.
    Currently ongoing atDHI • Working MPC - surrogate modelling framework within MIKE WORKBENCH • Demonstration of framework for reservoir flood control and optimisation of large-scale irrigation system • First tests of control of urban storm and wastewater systems © DHI 2. MPC 1. RR 3. HD-HiFi 0. Rainfall- Forecast #28
  • 27.
    Water Smart Citiesproject 2016-2020 • MPC-surrogate modelling framework − Extension of surrogate model class − Extension of methods for handling forecast uncertainty (incl. short to medium range weather forecasts) − Adaptive control automatically shifting between different control strategies depending on system state and rainfall forecast − Combined control of drainage system and WWTP • Forecast models − Probabilistic forecasting − Use of surrogate models − Data assimilation • Automatic surrogate model builder − MIKE URBAN -> surrogate model suitable for different purposes (e.g. forecast, control) • MIKE OPERATIONS implementation © DHI #29
  • 28.
    Thank you Dr. LisbethBirch Pedersen lpe@dhigroup.com © DHI #30