Smart Urban Water Systems
- some preliminary thoughts
Zhiguo Yuan AM, FTSE, IWA Distinguished Fellow
ARC Australian Laureate Fellow
Director, Advanced Water Management Centre
27 Feb 2020
Why do I feel passionate about it?
CRICOS code 00025BPresentation Title | Date 2
My Bachelor (1985)Sensors & instrumentation
My PhD (1992)Automation (AI)
Beijing University of Aeronautics and Astronautics
My career to dateUrban water management
Ghent University & The University of Queensland
Smart urban water systems
Knowledge-based fault diagnosis in dynamic systems (Yuan, Z. PhD thesis, 1992)
Smart Urban Water Systems
CRICOS code 00025BPresentation Title | Date 3
IoT sensors
Network communication
…
Sensing
Local control
Transmissions
…
Real-timecontrol
Learned status
Recommended strategies…
Data storage
Data analytics
Optimization
…
Computing power
Machine learning
Artificial intelligence
…
Supervisorycontrol
Planners
Managers
Operators
Customers
Operational/managing/planning decisions
Smart water system vs. ICA
CRICOS code 00025BPresentation Title | Date 4
• Smart water system encompasses ICA
Smart water
system
ICA: Instrumentation, control and automation
Smart water system vs. ICA
CRICOS code 00025BPresentation Title | Date 5
• Smart water system encompasses ICA
Smart water
system
ICA: Instrumentation, control and automation
but expanding towards
• higher levels of duties- More integrative - More supervisory- Decision support (operational,
management and planning, behaviour change)
• higher level of intelligence- IoT sensors- Big data- Machine learning & artificial intelligence
Technology push
• Internet of Things
• Low-cost sensors and low-cost, yet powerful
microchips
• Large data storage capabilities
• Fast computing e.g. iCloud computing
• Advanced data analytics
- machine learning
- artificial intelligence
6Presentation Title | Date CRICOS code 00025B
Driving forces for smart urban water
Demand pull
• Population growth and continued
urbanisation imposes more pressure to
urban water systems
• Climate change causes more variability
• Aging infrastructure
• Customer expectation
• Integrated urban water management is
becoming more important
Where are the opportunities (non-exhaustive)
CRICOS code 00025BPresentation Title | Date 7
Treatment
Domestic
use
Storm
water
Treatment
Treatment
Treatment
Surface
Water
Seawater
Ground
water
Industrial
use
Other
uses
SewersReceiving
waterWWTP
Distribution
TreatmentRecycled
water
CSO/SSO
Smart metering to
understand and change
consumer behavior
Integrated wastewater management
1. Sewer, WWTP and receiving water
2. Central and decentralized systems
Other possibilities
1. Business process automation
2. Interactions with customers
Example 1: water source optimization – the orange council case study
CRICOS code 00025B 8
Rainfall/runoff
Evaporation
Users
Shearing Shed bore
Stormwater schemes (SW)
Spring Creek Dam
Macquarie River
Showground bore
Bore 5 Batch pond
Suma Park Dam
Holding pond
Blackmans SW
BrooklandsSW
Somerset SW
MitchellCargo
SW
Escort SW
Env. flows
Env. flows
Env. flows
Env. flows
Losses Losses
Env. flows
Water Treatment
Plant
See Fig. 53 for details of
model downstream of Suma Park
Dam
Several water sources
(natural catchments, groundwater,
stormwater harvesting and
Macquarie River)
Several constraints
(environmental flows,
source withdrawal
limits, water restrictions)
Several objectives
(minimisation of costs, spill,
greenhouse gas emissions and
maximisation of environmental flows)
Presentation Title | Date
© CRC for Water Sensitive Cities
Dr Lisa Blinco, Uni Adelaide
Example 2: water production based on demand prediction
CRICOS code 00025BPresentation Title | Date 9
OUTIN
L
Controller /Switch
Set-point
Production flowDistribution
flowReservoir0%
20%
40%
60%
80%
100%
0
2000
4000
6000
8000
10000
22-Apr 00:00 22-Apr 06:00 22-Apr 12:00 22-Apr 18:00 23-Apr 00:00
m3 /
h
System #4, level based control
Distribution Production Reservoir level [%]
OUTIN
L
Prediction- and control algorithm
Set-point
Production flowDistribution
flowReservoir 0%
20%
40%
60%
80%
100%
0
2000
4000
6000
8000
10000
12-Apr 00:00 12-Apr 06:00 12-Apr 12:00 12-Apr 18:00 13-Apr 00:00
m3 /
h
System #4, predictive control
Distribution Production Reservoir level [%]
Flow prediction
model learnt from
historical data
Bakker et al. (2013)
Example 3: tackling non-revenue water
CRICOS code 00025BPresentation Title | Date 10
Example from Unitywater (QLD) potable water distribution network using TaKaDu platform (Israel):
• Real-time flow data versus Network-based prediction flow model = leak detection shortly after burst & detection of hidden
(underground) leaks
• $16 million AUD per annum in savings from early detection of leaks
• 6.5 billion litres of non-revenue water loss prevented in 2017 due to enabling of rapid, fit-for-purpose and localised response
Example 4: water consumption pattern analysis – changing customer behavior
CRICOS code 00025BPresentation Title | Date 11
One-off exceptions
accounted for 31% of
all water use
© CRC for Water Sensitive Cities
Prof Rachel Cardell-Oliver, UWA
14,000 smart meters installed
City of Kalgoorlie-Boulder
Example 5: online optimization of pump operations
Odour control at inlet of a WWTP
13 pump stations; 26 km
ADWF = 14.3 ML/day
Mg(OH)2 ⇌ Mg2+ + 2OH-
H2S ⇌ HS- + H+
Magnesium dosing
PS2 WWTP
PS3
PS4
PS6
PS7
PS8
PS5
3000 m
Φ0.4
m
162 m
Φ0.325m
218 m
Φ0.6m
100 m
Φ0.3
m
259 m
Φ0.45m
356 m
Φ0.2
25m
1441 m
Φ0.45m
498 m
Φ0.3
75m
1484 m
Φ0.45m
225 m
Φ0.225m
787 m
Φ0.3
25m
477 m
Φ0.3
75m
873 m
Φ0.525m
372 m
Φ0.1
m
544 m
Φ0.525mCP1 CP2 CP3 CP4 CP5 CP6
PS1
1822 m
Φ0.5
25m
TH5
TH3
88 m
Φ0.2
25m
TH2
57 m
Φ0.1
m
TH1
537 m
Φ0.1
5m
C30
1215 m
Φ0.3
3m
C31
7 m
Φ0.1
5m
2037 m
Φ0.525m
470 m
Φ0.525m
1100 m
Φ0.525m
200 m
Φ0.525m
200 m
Φ0.525m
400 m
Φ0.525m
Treatment
plant
Odour control at inlet of a WWTP
• Hybrid system – linear model predictive control algorithm not directly applicable
• Large search spaces – efficient optimization algorithms required
PS2 WWTP
PS3
PS4
PS6
PS7
PS8
PS5
30
00
m
Φ0
.4m
162 m
Φ0.325m
218 m
Φ0.6m
10
0 m
Φ0
.3m
259 m
Φ0.45m
35
6 m
Φ0
.22
5m
1441 m
Φ0.45m
49
8 m
Φ0
.37
5m
1484 m
Φ0.45m
225 m
Φ0.225m
78
7 m
Φ0
.32
5m
47
7 m
Φ0
.37
5m
873 m
Φ0.525m
37
2 m
Φ0
.1m
544 m
Φ0.525mCP1 CP2 CP3 CP4 CP5 CP6
PS1
18
22
m
Φ0
.52
5m
TH5
TH3
88
m
Φ0
.22
5m
TH2
57
m
Φ0
.1m
TH1
53
7 m
Φ0
.15
m
C30
12
15
m
Φ0
.33
m
C31
7 m
Φ0
.15
m
2037 m
Φ0.525m
470 m
Φ0.525m
1100 m
Φ0.525m
200 m
Φ0.525m
200 m
Φ0.525m
400 m
Φ0.525m
• Controlled case: 15% of time below target
• Uncontrolled case: 38% of time below target
Example 5: online optimization of pump operations
Machine learning
CRICOS code 00025BPresentation Title | Date 14
• Block-box (data driven)
⎼ Statistical modelling
⎼ Artificial neural network
⎼ Large amount of data
⎼ Identifiability issue
( ) ( ) ( ) ( ) ( ) ( )1 1 1A z y t B z u t C z e t− − −= +
ARMAX model:
Processu(t) y(t)
Machine learning
CRICOS code 00025BPresentation Title | Date 15
• Block-box (data driven)
⎼ Statistical modelling
⎼ Artificial neural network
⎼ Large amount of data
⎼ Identifiability issue
ARMAX model:
( ) ( ) ( ) ( ) ( )
( ) ( ) ( ) ( )
( ) ( ) ( )
0.2140 1 0.1822 2 0.1018 3 0.1634 4
0.1014 5 0.0869 6 0.0543 7 1.5261
ˆ ˆ ˆ ˆ
1
0
ˆ
.7946 2 0.0285 3 0.
ˆ ˆ
292
ˆ
1 4 ( )
y t y t y t y t y t
y t y t y t u t
u t u t u t e t
= − + − + − + − +
− + − + − + − +
− − − + − +
Li et al. (2019)
Machine learning
CRICOS code 00025BPresentation Title | Date 16
Machine learning
CRICOS code 00025BPresentation Title | Date 17
• Block-box (data driven)
⎼ Statistical modelling
⎼ Artificial neural network
⎼ Large amount of data
⎼ Identifiability issue
• Grey-box
⎼ Data supplement existing knowledge
⎼ Model-supported data analysis or data-enabled model identification
⎼ Process knowledge required
⎼ Smaller amount of data needed
Prior knowledge is important
We should not ask the machine to invent the Bernoulli equation!
Modelling of methane product ion in gravity sewers
A Simplified M odel for M ethane Product ion in Gravity Sewer
Methane product ion rate in a gravity sewer is proport ional to the area of biofilm that is in contact
with wastewater. For a given wastewater flow (Q), pipe diameter (D) and slope (S), the biofilm
area can be theoret ically calculated as follows (Akgiray, 2004).
Figure 1: Schemat ic of water flow in a sewer pipe
✓=3⇡
2
vuuuut 1−
vuuut 1−
vuut
⇡Dn
D83 S1
2
(1)
Where Q is flow rate in m3/ s, n is the pipe frict ion factor, D is pipe diameter in m, and S is the
slope.
The above relat ionship can be represented by a simplified expression as follows:
✓= k ·Q↵·D β ·Sγ (2)
Theoret ical ✓values were est imated using Eqn 1 for a series of pipe size with diameter ranging from
100 mm to 3000 mm, and pipe slope ranging from 0.0005 to 0.02. The correlat ion between the ✓
values thus calculated and those est imated using the simplified equat ion (Eqn 2) after determining
the parameters through regression is presented in Figure 2.
Area of biofilm can thus be calculated using the simplified expression as follows:
Abf = ✓·D
2·L = k ·Q
↵·D β ·Sγ ·
D
2· L (3)
Abf = k ·Q↵·D β ·Sγ · L (4)
Where L is length of pipe in m.
Methane product ion rate can then be est imated as
r CH 4= k
0
·Abf = k ·Q↵·D β ·Sγ · L (5)
Where r CH 4is the methane product ion rate in kg CH4/ day.
Considering 1 km length of pipe and the temperature of 20oC as standards, methane product ion
rate can be expressed as follows:
r CH 4 ,20 = k ·Q↵·D β ·Sγ (6)
February 1, 2016 Page 1
Machine learning
CRICOS code 00025BPresentation Title | Date 18
• Block-box (data driven)
⎼ Statistical modelling
⎼ Artificial neural network
⎼ Large amount of data
⎼ Identifiability issue
• Grey-box
⎼ Data supplement existing knowledge
⎼ Model-supported data analysis or data-enabled model identification
⎼ Process knowledge required
⎼ Smaller amount of data needed
𝑘𝐶𝐻4,𝑇 = 𝑎1𝑇𝑏1 + 𝑎2𝑄
𝑏2 + 𝑎3𝐷𝑏3 + 𝑎4𝑠
𝑏4
𝑘𝐶𝐻4,𝑇 = 𝑘 𝑇 𝑄𝛼𝐷𝛽𝑠𝛾
D: diameters: slopeQ: flow rateT: temperature
Segmented efforts need to be united
CRICOS code 00025BPresentation Title | Date 19
WHAT UTILITIES WANT TO ACHIEVE?
(DEMAND PULL)
WHAT DO TECHNOLOGIES OFFER?
(TECHNOLOGY PUSH)
SHARED
OBJECTIVES
JOINT
EFFORT
• Urban water system will get smarter!!
• While isolated case studies have been done, there is a lack of a systematic
framework
• Collaborative efforts from utilities, hardware suppliers, software suppliers,
and researchers needed
• Multi-disciplinary research is required
Concluding remarks
CRICOS code 00025BPresentation Title | Date 20
Acknowledgements
CRICOS code 00025BPresentation Title | Date 21
Smart urban water system workshop
Brisbane, Nov 26, 2018