Systems Engineering for Water Management A decade of Water - - PowerPoint PPT Presentation

Systems Engineering for Water Management

A decade of Water Information Network collaboration

Outline

• Water management?
• An information infrastructure
• From data to model
• Control
• Ongoing work & challenges

• “Water management crisis”

– Water efficiency < 50% – 55 to 60% of “easy” water is in use

• Water consumption pressures

– Equity – Industrialisation – Irrigation expansion (70%

• f all water usage)

– Climate change – Population growth

• No change scenario, the

world runs out of “water” in 2025

UNESCO World Water Reports 2003-2005-2009

• >30% of extracted water is

unaccounted for

• 1B people have access < 10l/day

2B people have access < 50l/day

• human activity → climate change
• Economic & physical water scarcity

National Geographic

Lake Chad 1972 Lake Chad 2007

UNESCO World Water Report 2: Water scarcity estimate China, India, USA, Australia all face serious challenges

Climate change in action? Human impact in action

Victoria= 1/3 Texas Murray-Darling Basin = 1.5 Texas

• Efficiency < 50%
• Over-irrigation leads

to soil degradation

• Accountability?

Dam evaporation ≈ 8% Dam release 100

Plants store (1%) 0.4

• Low energy footprint
• More productive land
• 50% of all farm profits

Channel to farm consumes 30

• Seepage ≈ 5
• Evaporation ≈ 5
• Outfalls ≈ 5
• Conservative

management ≈ 15

• Outfalls ≈ 15
• Seepage ≈ 15
• Plants ≈ 40

Metering error ± 20%

Farm gate to plant consumes 30

Manual on site operation Ordering delay > 3 days Poor regulation (30cm fluctuation in water level) OHS issues

No 25 No 49

No 35

06/12/2001 02/05/2002

24h between adjustments Drop bar structure Manual overshot gate Dethridge Meter Wheel +/-20% accurate

Manual undershot gate to farm

Summary

• Water is a limiting resource
• Water usage efficiency <50%
• Irrigation accounts for 70% of all water usage
• Irrigation

– gravity fed open channel distribution system – (semi)-manually operated (reservoir-channel-gate) – policy based, open loop, exploitation regime – poor information (infrastructure) – materially unchanged since the hanging gardens of Babylon (700BC)

A pool, gates, datum

1 , − i u

y

i u

y ,

1 + i

g

i

g

i d

y ,

1 , + i d

y

Datum level may not be unified across system

Canal slope ≈1/10,000

Water level for no flow Pool = canal section between gates Length varies from 1km – 10km Pivot point Off take Off take close to pivot point

i

ν

1 − i

ν

i

u flow

1

flow

+ i

u

• Water tight
• Self cleaning
• Low head loss
• Precision manufacturing

– Accurate & repeatable measurements – Flow actuator & meter – Precision control – High duty cycle

• Radio based internet
• 1 PC on board
• Patented technology

The FlumeGate™

On-farm ad-hoc network

Information Infrastructure

• Water levels, gate positions, flows

at all regulators and farm off takes

• Soil moisture, plant response

and on-farm irrigation actuation

• Radio internet along channel

(hop number <8)

• Ad-hoc network on farm
• Event based sampling & actuation
• Not fully integrated across farm

gate (demonstrators only)

2.4kb/s links F1 2.4kb/s links F2 Repeater 19.6kb/s 4 Fs 10-30km ≈100km Line of sight 1-3km (<10km) On-farm ad-hoc network

Water managed from reservoir to plant Information feedback loop from the “crop” perspective moderated by the “overall system” perspective

Water Information Network

On farm experimental Channel System Commercial Gate Regulator main Central node Farm nodes

Central Goulburn 2004-2006 New ICT infrastructure (retrofitted) Gate (design & calibration) patented Automated

• peration, both
• n canal and

to farm Water level & flow monitoring & control

Minor canal Main canal To farm Replaces Dethridge wheel

Information Infrastructure

• Target water efficiency & on-farm profitability

– Reliable, quantitative information – Real time water balance creates a water market for buying and selling of water in real time – Water-on-demand – Reduce ordering delay (better on-farm management) – Improve water level regulation (better farm land command)

• Build a data based dynamic model

– Enable short term prediction (say over a week) – Enable feedback based control – Inform long term policy

• Scale, retrofitting, expandability

Modeling

• Pool = from up-stream flow to down-stream water level,

with a downstream flow disturbance

• View overall system as a concatenation of such models

(generic structure for all flow distribution systems

• Inputs u, outputs y, disturbances v

Pool i-1 Pool i

i

u

i

y

1 − i

u

1 − i

y

1 − i

ν

1 − i

d

+

i

ν

i

d

+

. . .

• . . .

The Art and Science of Modeling: from data to model

Model complexity Utility (short term prediction) Data complexity

The aim of modeling: For a given set of data (= data complexity), there is an optimal (= best utility) model complexity

• ptimal data complexity –
• ptimal model complexity

St-Venant models

( )

) ( ) ( ) ( ) ( ) (

1

t t u t u k t y p

i i i i i i

ν τ σ − − − =

+

Inflow

Grey Box Modelling

Outflows

i

i

pool across delay τ ( )

i

y t ) (

1 t

ui+

1( ) i

y t

−

) (t ui

Off-take

• n pool i

World-wide patent granted

mass balance + waves = (leaky) integrator + lightly damped pole pair Inflow Outflow

) (t

i

ν ) ( ) ( geometry f t u

i i

=

St-Venant Equation Model

• In 1871 a 74y old St-Venant proposed a PDE

(now called a 1D-Navier-Stokes) expressing conservation of momentum & mass with (viscous) friction in un-steady flow

• Boundary conditions = hydraulic characteristics of

gates (need system identification techniques, scale issues!)

• Must identify “friction” and “geometry” from data to

make a predictive model (non-linear terms) (hard work)

• 100+ year old model; first “real canal” models tested in

2000; first labo-canals tested/verified in 1971 Brutsaert

1797-1886

= ∂ ∂ + ∂ ∂ x Q t A

( )

2

2 2

= − + ∂ ∂ + ∂ ∂         + + ∂ ∂ S S gA x Q A Q x A A Q B gA t Q

f

take

• ff
• utflow,

and inflow for conditions Boundary slope bed mean and slope friction gravity surface water at channel

• f

Width Flow area sectional Cross − − − − − S S g B Q A

f ,

Continuity equation along the channel Momentum equation along the channel

200 220 240 260 280 300 320 340 23.89 23.91 23.93 23.95 23.97 23.99 24.01 24.03

Time (min.) Water level (mAHD)

First order Third order

PDE (with estimated parameters) Validation data set

Simulated models, compared with data 3 pool canal section Wave period ≈ 10min Delay ≈ 3.3min 900m downstream

Modelling for Automation

• Grey box models are preferred

– 3rd order model suffices ( <10km pools) – Models validated across complete flow regime over 4 irrigation seasons (set point regulation!) – Model structure validated, easy to tune against data, both in closed and open loop identification mode – Physically relevant parameters easily recognised (delay, dominant wave frequency, dominant time constant) – Scales well

An Open Question in Model Order Reduction

a) nonlinear PDE models b) closed loop validation water level regulation, guided model validation 3rd order grey box model + regulator characteristics St Venant equations + regulator characteristics

Data (flow and levels) Model order reduction

?

Summary Model

• Simple input/output models suffice
• “Grey box model”

– Water balance, waves, boundary conditions (gate characteristics), delay time – Simulates well, over extended periods of time (week)

• Models can be tuned from operational data

Decentralized Structure Preserving Control

Regulation, rejecting disturbances, suppressing waves NON-TRIVIAL CHOICE of PAIRING VARIABLES

. . .

Pool i-1

1 − i

u

1 − i

y

1 − i

ν

+

• Control

at gate i-1

+

1 − i

r

• Pool i

i

u

i

y

i

ν

+

• Control

at gate i

+

i

r

• . . .

1 − i

d

i

d

TCC™

Pool i Local datum line

FB FF

downstream flow demand Water level error

Local

Downstream Gate reference

• Filtering waves
• Gate characteristic

inversion

• Anti-windup

Off-take

IFAC World Congress 2005 29

Total Channel Control™

• Total Channel Control™ industrial size implementations since

2002, over 2000km of canal in operation

• Improvements

– Distribution efficiency ≈ 90% up from ≈70% (CSIRO audited) – Copes easily with start/stop events (rain)

• Other benefits

– Water leak detection – Water flow accounting; (balances to about 2% accurate); enables real time water market – Water application totally different (farmers adapt, and obtain better on-farm outcomes) – Better regulation implies a higher water set point is feasible = more (flow) capacity and more land that can be irrigated within the same infrastructure

TCC™ Limitations

Set point tracking demands

Local controllers require global optimization

( )

i i i s i i

u u e s G y

i

ν

τ

− − =

+ − 1

) (

) )( ( ) (

1 i i i i i i

r y s T u s F u − − =

+

... ) ( ) ( 1 ) ( ) ( ) (

1 +

+ + = ⇒

+ − i s i i i i i i

u e s T s G s T s G s F u

i

τ

Upstream input propagation operator Arbitrary pool dynamics Any control

1 ) ( ) ( 1 ) ( ) ( ) ( = + +

i i i i i

T G T G F

) ( ) ( 1 ) ( ) ( ) ( > = + + ⇒

= − i s s i i i i i

i

e s T s G s T s G s F ds d τ

τ

• Upstream inputs are amplified
• Sooner or later the inputs will

saturate!

IFAC World Congress 2005 31

TCC™ Response (Simulation)

1000 2000 3000 4000 21.05 21.1 21.15 21.2 21.25 Time in min Water level in m

Water level Set point Disturbance response

• 4

4 8 5 10 15 25

Time (h)

20 55 (22) 115 (46) 75 (30) 95 (38) Down stream water level (cm) Demand Ml/day (cfs)

287 “measurement events” ≈ 5min 1 gate movement / 10min Pool with the largest water level error (first pool in the channel) 1st pool on CG2 01.01.2005

• 8

Controller Issues – Event based?

• Event based sampling

– None of our “techniques” apply directly – Minimizes telecommunication load on network

• Ad hoc approach used, event based sampling is resampled =

normal sampling + bounded error = don’t care?

• Question: what does optimal control under event based

sampling actually look like?

• Question: what is the minimum data rate?

1 2 3 4 5 6 7 8 9 10

• 0.5

0.5 1 1.5 2 E E E E E E E E

Controller Issue - Error Propagation?

• Exponential error propagation = limits the size of the network

It is a direct consequence of the decentralized controller choice + tree-like network graph, it is independent of linear dynamics as long as steady state condition is met

• Ad hoc solution, is to cut the network into sub-networks, and

create a hierarchy of controllers

• Question: are there other network topologies, or different controller

choices, for which the error propagation is bounded (regardless of the size of the network)? (Conjecture = No, for linear dynamics)

• Question: does the conclusion for decentralized also hold for

nonlinear controllers (Conjecture = yes, but may be not exponential growth)

Summary Control

• Control achieves good regulation

– ± 10cm water level (at worst), more like ± 2cm – meets demand in near real time – avoids waves

• Practical issues that remain

– scale & hierarchy over space & time (factor 100?) – incorporating other feed forward information: demand prediction & weather forecast (MPC ?)

• Academic issues that remain

– Topology selection, error propagation, event based sampling

Water managed from reservoir to plant Information feedback loop from the “crop” perspective moderated by the “overall system” perspective

Water Information Network – on Farm?

On farm experimental Channel System Commercial Gate Regulator main Central node Farm nodes

Horticulture 10% of all water Water Use ML/ha Yield t/ha Gross Margin A$/ha Economic Water-Use Efficiency A$/ML $ Return

(after buying water)

A$ per extra ML Manual 2.81 30 7,500 2,700

• On demand

5.56 31 25,000 4,600 6,100 Comparison +100% +3% +233% +70%

• Dairy with flood

irrigation 50% of all water Total volume

• f water

consumed per season Productivity Gross margin Additional Income (A$500/ML) (ML/ha) (ton dry matter/ML) (A$/ha/yr) (sale of water) Manual 11.25 1.3 550

• On demand

8.27 1.8 760 1,710 Comparison

• 30%

+40% +40% +300% TCC™ enabled on farm automation + market transfer = pays for itself in 1 season

Further challenges

• Scale up over an entire river basin

– Integrate demand & supply (water market, short term weather forecasts, control demand?) – Scale by another factor of 10-100 both in time and space (long term forecasts) – Integrate all use: rural & urban/industrial & ecological – Integrate entire water cycle : harvesting, natural & engineered distribution, surface & ground water, drainage, re-use … – Integrate water quality

Thank You & Questions