How many water utilities have a dedicated data scientist in their ranks? Probably not as many as they should. Here Dr Peter Prevos from Australian water company Coliban Water speaks to Tom Freyberg about why he’s cautious about machine learning and his aspirations to be the ‘Willy Wonka of water’.
Embracing data science magic in Australia
If you’re asked to name a famous magician you might probably think of Harry Houdini, David Copperfield, David Blaine or even American entertainers, Penn & Teller.
Escaping from a water torture cell, making the Statue of Liberty disappear, levitating and catching bullets all earned these illusionists a place in the magicians’ hall of fame.
Now if you’re asked to name a famous magician and data scientist in the global water market, it’s unlikely you’ll think of anyone.
Yet one man who hopes to change this, to become a household name in data science and who, in his own words, “aspires to be the Willy Wonka of the water industry” is Dr Peter Prevos, data scientist from Australian utility Coliban Water.
Bridging IT and OT
While many other Australian utilities don’t have a data science manager, “although larger utilities have specialised data scientists”, Prevos says Coliban Water is different in that it has a dedicated data science strategy.
In his words, this strategy breaks down data into three different aspects:
- The operational technology (OT) team – the Supervisory Control and Data Acquisition (SCADA) team who collect the data
- IT team – responsible for storying the information and making it available to everyone
- Data science team – helping the organisation to create value from that data.
“Because the IT manager is not a water professional, they're not really that involved with what you do with the data,” he explains. “The OT team are more interested in making sure that the data is available all the time so that assets keep operating. My interest is creating value from data by looking at problems holistically by combining different data sets. One example is combining customer feedback with engineering data.”
Smoking computers and data points
Coliban Water manages AUS$1.56 billion of assets, including 2333km of water mains, 504km of sewer mains, 16 treatment plants and 177 wastewater pumping stations.
With the development of a data science strategy, it raises the question of what benefits, or operational changes have been made in light of Prevos’ data analysis?
The Water Services Association of Australia has produced a manual to assess the performance of water treatment plants. Part of this manual contains a mathematical assessment of treatment plant performance.
“When we tried to do that with a spreadsheet, virtual smoke came out of the computers because they're not designed for that sort of workload,” says Prevos. “You're analysing millions and millions of data points, it's really nudging into the big data space.”
His team set out to develop a system to “automatically clean and analyse SCADA data”. Prevos says these results can now be produced in minutes, in “an automated way”. Instead of people having to trawl trough data, the system sends emails weekly, monthly or whatever the time frequency is required.
Not content with the improvements in workflow and time improvements to generate reports, Prevos and the team are still undergoing “a lot of data cleaning” involving “going through all data with a fine tooth comb and trying to work out how to interpret the information”.
“It's a journey in discovery, more than anything else,” he muses.
The algorithms behind traffic lights
Another success saw Prevos and his team translating monthly water quality reports for the board of directors from “technical information that could be difficult to understand” into a simple traffic light system.
This required a lot of experimentation to develop the algorithms. “You can only do that if you have somebody who has the time and the capability to write the code … underneath these simplistic reports there’s an enormous amount of complexity,” he adds.
To bridge the gap between IT and OT, requires water engineers with a deep understanding of data science. This is becoming a much-needed skillset and engineers evolving into this space are becoming highly valued.
Prevos adds: “I've learned a lot about the information technology part, so I can talk shop with the IT team and the OT team, and then I can say we can solve this problem in this way because we have these data sets."
A migration down under
Leading up to his current role as manager data science, Prevos has had an interesting and varied career.
A Dutch engineer graduate from Zuyd University of Applied Sciences in the Netherlands, Prevos spent time at Boskalis as a project engineer supervising and managing coastal defence projects from Netherlands, to England, Hong Kong, Singapore and Bangladesh.
After a short stint at Dutch government administration company Rijkswaterstaat, he moved to Australia in 2000 “for my wife, Sue”.
It was in 2000 when he joined Coliban Water, starting on project management before eventually moving up the ranks into data management. “This is my seventh job title – I think I need to stop putting titles on my business cards because they keep changing”, he points out.
He went on to study arts and philosophy at Monash University before taking an MBA and a PhD in business from La Trobe University.
It was back in 2013 when the organisation needed someone to “look after data issues we had in the operational space. Back then it was called systems monitoring because it had an operational focus”. Since then Coliban Water moved to a data science approach that encompasses the whole organisation.
“I took the job because I had an interest in data and an affinity with computers”, he adds.
Improving water utility data culture
Prevos sees his role in not just making sense of masses of volume of data but also helping to improve what he calls “data literacy and culture” from within: essentially educating non-data people about how to best use and interpret data.
“It's about the people and giving them the ability to understand that information, the ability to understand the reality that it comes from,” he explains.
“The part I want to play in the industry is help other water professionals to go through a similar journey that I had, to start learning to write computer code and create that data literacy from the bottom up, rather than buying the latest gadget, machine learning algorithm that few people understand.”
Rationalising the digital hype on water
Related to this, the data scientist says he loves to hate the phrase “digital water utility”, which he believes is often overused and used unnecessarily.
“It’s easy to carpet bomb a region with Internet of Things (IoT) sensors and say: “Yes, ‘we’re a digital water utility’ but what is being done with all that information?”
Prevos is also keen to rationalise much of the digital hype when it comes to water.
“People talk about digital disruption but this only exists when your core service is being disrupted and replaced by a digital offering,” he says. “That’s never going to happen to a water utility. Digital technology is not going to change the substantial reality which we deal with: water weighs a tonne per cubic metre and we have to push that uphill.
“No amount of digitization is going to change the fact that water will be scarcer. It will help us surely to manage these problems better, but we're never going to be in a situation where ones and zeroes come out of the tap. So the digital is only a tool for us to be a water utility.”
While Prevos respects new tools including machine learning, at the same time he remains cautious.
He refers to one example of an algorithm was able to predict patterns extremely well. “It’s quite spooky what it can do,” he observes.“ The method was, however, very complex and rejected by the managers because it was a “black box”. This technology will allow us to better understand our systems, but we need to ensure we educate ourselves on how to best use these new methods, Peter reflects.
“Don’t be creepy”
As well as his day job at Coliban Water, Prevos is also an author, having recently published the Principles of Strategic Data Science.
Despite the complex algorithms and coding that goes into data analysis, when discussing the uses for customer data, Prevos breaks it down to one simple ethical question: is it creepy?
“The first rule of data science ethics is don’t be creepy,” he laughs. “For example, in our digital metering reports, we've locked on all reports that show information about individual customers to my colleagues, unless they have a real dying need to look at an individual customer. Because you can find out a lot about people's lifestyles, and there's no reason for that.”
He’s a firm advocate of “data science ethics” which he also calls “algorithmic justice – the way you analyse the data shouldn’t be used to disadvantage your customers”, he says.
Dangers of data deception
On the topic of data magic, Prevos used to perform magic shows for children’s parties but never wanted to further it as a commercial venture. “I was no longer motivated to enter some stranger’s home to entertain their kids, so eventually lost interest,” he admits.
Today, he retains an interest in magic and writes books about science and magic using his professor account, the ‘Lucid Manager’. He even recently threw in a few magic tricks for a presentation about innovation at Coliban Water.
Yet the experience in magic in the past still benefits Prevos’ present role and his ability to judge between people using data for deception.
“Data science can be very deceptive and that’s where it overlaps with magic,” he says. “A magician is very honest because they tell you that they will deceive you and then they do it. But data deception is probably more hidden. It can be unintentional if you don’t understand, if you’re not data literate it is easy to make mistakes and draw the wrong conclusion.”
A shining light in the dark data era
The quote “Data is the new oil” can be credited to mathematician Clive Humby. This comment picked up steam after the economist published a 2017 report entitled “The world’s most valuable resource is no long oil but data”.
Despite its potential, data has also earned itself a bad reputation, spawning the phrase ‘dark data’. Take the Facebook/Cambridge Analytica scandal, or reports suggesting home AI systems are listening and storing data from daily conversations, handpicking keywords only to target those potential customers with targeted advertising on other connected platforms. There are multiple examples.
People are becoming naturally wary of data and online privacy, particularly in a post-GDPR era. As a result, water utilities have a level of responsibility when it comes to protecting and using consumers’ data wisely.
The larger than life Peter Prevos has cemented his role as data scientist; a job that will surely become much more common in the global water utility space. Clearly able to turn his hand to most challenges, he’s carving out this niche and making it his own.
He jokes that he aspires to become the “Willy Wonka of water”, inspired by Roald Dahl’s fictional character who he sees as a creative innovator who defies the laws of tradition.
Perhaps this character doesn’t really do justice to the Coliban Water scientist. After all, both the Gene Wilder and Johnny Depp versions of the Roald Dahl character had their darker sides. Instead Prevos is determined to keep everything clean about data: the ethics, the quality, the analysis and results.
While he may not be a household name of Harry Houdini, David Copperfield or David Blaine of water data just yet, believe me, he’s only just getting warmed up.
Dr Peter Prevos will be hosting a Data Dive workshop at the Aquatech Innovation Forum on November 4 in Amsterdam, part of the wider Amsterdam International Water Week. For more information, email: firstname.lastname@example.org