Transcript Water Sensitive Cities - The University of Western Australia

Intelligent Urban Water Systems
Jin Wang
The University of Western Australia
16-09-2014
INDUSTRY CONTEXT AND PROBLEM
ENVIRONMENT
CRC for Water Sensitive Cities
• CRCWSC researches urban water reform required to
transform our cities into liveable, sustainable and productive
cities.
• Our research over the next nine years will guide capital
investments of more than $100 Billion by the Australian
water sector and more than $550 Billion of private sector
investment in urban development over the next 15 years.
© CRC for Water Sensitive Cities 2012
Smart Water Metering
Smart meter = asset (install, maintain)
Smart metering = discovery of knowledge
to support decision making
Customers – empowerment, informed behaviour
Policy – billing bands, future investment, planning
Environmental – reduce water use, carbon footpr.
Operational – reduce OHS costs, delay/avoid new
infrastructure
Intelligent Urban Water Systems
Goal
To develop techniques for utilizing sensor data to
optimize the efficiency and safety of urban water
systems
Areas
1. Data mining of patterns from smart water meters
2. Optimization of pumping from multiple
alternative sources of water
3. Optimizing sensor selection and placement for
meeting information goals e.g. detecting leaks in
pipeline systems.
Smart Water Meter Data
500 1000
0
Litres per hour
Household meter readings, average = 590 l/hh/d
0
1000
2000
3000
4000
5000
Hours of Trial
800
400
0
Litres per hour
Household meter readings, average = 1115 l/hh/d
0
1000
2000
3000
Hours of Trial
Raw data demo: Raw data
4000
5000
RESEARCH AIMS AND QUESTIONS
INDUSTRY AIMS AND QUESTIONS
Data mining
Data mining is the process of discovering
interesting patterns and knowledge from large
amounts of data
Modes of enquiry:
• explore (what)
• explain (how)
• predict
• plan
Data mining = searching for interesting patterns
in data in order to:
• Characterize and discriminate categories
• Identify frequent patterns, associations and
correlations
• Predict future behaviour using classification and
regression
• Discover clusters in labelled and unlabelled data
• Analyse outliers and detect anomalies
[Data Mining, Han, Kamber & Pie (2006)]
Data mining example- characterize and
discriminate
• Example from smart metering
– Continuous flow patterns were prevalent in the
Kalgoorlie sample with 84% ( 157 / 188 ) of
households having at least one day of continuous flow.
Continuous flows accounted for 10% (3 / 29
megalitres) of all water used by houses in the sample
population.
Data mining example-identify frequent
patterns, associations and correlations
• Example from smart metering
– For meter 99 on 50 / 170 days recorded water use was
relatively high, totaling 15 megalitres (30 %) of 99’s
overall water use. This high water use occurred most
frequently on Mondays and Fridays, between 6am and
12 noon on those days.
Q 1 : What types of water use occur in
Kalgoorlie? Karratha?
Perth? Melbourne? Brisbane?
(explore)
Industry problem:
Discover “unknown unknowns” of water use
Are the assumptions behind water saving campaigns
correct?
What are the new opportunities for customer
engagement?
Approach:
Build a conceptual data model for water-use activities
using (only) hourly observations
Q 2: Identify temporal patterns of peak
demands and the activities behind them
(explain + plan)
Industry problem:
Infrastructure planning: delay or avoid $M upgrades by
engaging with customers to modify or reduce their use
e.g. offset their watering schedules
Approach:
Data mining models to automatically query temporal
patterns in populations
Q 3 : Identify “inefficient” garden watering
activities. How much / when is water is used
this way?
(explain)
Industry problem:
Target and engage a small number of significant customers
e.g. those with “inefficient” watering habits. Are all inefficient
garden users also high overall water users? Is highly inefficient
garden use prevalent or isolated users?
Approach:
Searching for relatively rare patterns
(eg 1 hour per 168 hours in a week, needle in haystack)
Searching for complex temporal patterns
(e.g. every Mo,Tu,Sa at 2am, every even day of the month at 3am, twice a week on
different days, during summer but not winter)
CASE STUDY
Case Study 1: Kalgoorlie-Boulder WA
238 properties (of 13,800) 14 months
Raw Data
500 1000
0
Litres per hour
Household meter readings, average = 590 l/hh/d
0
1000
2000
3000
4000
5000
Hours of Trial
800
400
0
Litres per hour
Household meter readings, average = 1115 l/hh/d
0
1000
2000
3000
Hours of Trial
4000
5000
Q 1 : What types of water use occur in Kalgoorlie?
(explore)
Research Contribution: a conceptual data model for
water-use activities using (only) hourly observations
Findings (1): Anomalies are interesting
94% of
users have
continuous
flows
(leaks)
One-off exceptions on
accounted for 31% of
all water use
Findings (2): Temporal patterns are interesting
Regular high
use on Mon,
Wed, Fri but
only in summer
Case Study 2: Karratha WA
100 households, 111 days
Q 2: Identify temporal patterns of peak
demands and activities behind them
(explain + plan)
• Q: Is water demand related to land size?
A: Possibly yes (work in progress)
Q 2: Identify temporal patterns of peak
demands and activities behind them
Q: Would adjusting a few customers’ recurrent habits achieve smoothed
demand peaks?
A: Possibly yes (work in progress)
Q 2: Identify temporal patterns of peak
demands and activities behind them
Q: Would adjusting a few customers’ irrigation habits achieve smoothed
demand peaks?
A: Possibly yes (work in progress)
Q 3: Identify “inefficient” garden watering
activities. How much water is used this way?
When?
(explain)
• “Intentions to save water are shaped by attitudes,
beliefs, habits and routines, personal capabilities
and contextual factors”
[Russell S, Fielding K (2010) Water demand management research: A psychological
perspective. Water Resources Research 46(5)]
•
•
•
•
Human behaviour (calendar patterns) are complex
Human behaviours are non-stationary
Problem: how to detect habit patterns?
Aim: reliable, automatic generation of evidence for
decision making
200 400 600 800
0
Volume (L) per Hour
Finding patterns – where is the irrigation?
51 53 55 57 59 61 63 65 67
Week of Year
Water Use Habits (Irrigation Patterns 1)
Water Use Habits (Irrigation Patterns 2)
Irrigation
Actual use
136 kL
Budget 12 kL
INDUSTRY IMPACT AND BENEFITS
Industry Benefits (1): Water Use Sig Patterns
• Kalgoorlie trial was a response to safety for meter
readers (OHS)
• Independent report: “Smart meters have reduced
water use in Kalgoorlie-Boulder by reducing the
average duration of leaks [using] … suspected leak
letters…. saved an estimated 13,713 kL in 2012/13
• There remains significant scope to leverage the
information provided by smart meters
• Targeted customer engagement is critical”
‘Intelligent networks - Smart Metering and
Data Logging Programs Evaluation’, Report
for Water Corporation by Marsden Jacob
Associates Pty Ltd, January 2014
Industry Benefits (2): Water Use Habits
• Ranked list of customers for engagement
• Identified potential for savings from inefficient
irrigation – just the excess (53%)
• Configurable, automatic search supports
evidence-based decision-making
• Answers new queries:
– All high irrigators = all high water users ?
• Answers new queries:
– What are the temporal patterns for peak demand? Is
there scope to adjust ? (delay infrastructure)
Summary
• Developed a novel model of water use
signature patterns for medium-resolution
smart metering
• Pattern discovery and description is
automated, making the approach scalable for
large user populations over long time periods.
• Novel way to identify calendar habit patterns
• Supportive industry partnership – ongoing
work towards technology transfer
© CRC for Water Sensitive Cities 2012