When a regulator tasks a water utility to cut leakages by 10 million litres per day, there is a clear business case to find a solution.
This was the reality for UK water supply and treatment utility Yorkshire Water, Andy Sewell, Telemetry Manager, told the audience at OSIsoft’s User Conference in London.
Half of Yorkshire Water’s 83,000km of distribution network constitutes ageing cast-iron pipework circa 1950s. Digging holes is costly, disruptive and lengthy so the utility turned to the predictive power of data.
Sewell said at the event in October 2017 that the utility has adopted an early leak identification solution in a bid to meet three core objectives - cut operating costs, meet leakage targets and prevent customer outages.
As the cost of advanced sensors has dropped, the business case for smart water deployments is more solid. And as Gary Wong, Global Water Industry Principal at real-time data management company OSIsoft, said in an Engerati webinar, a water utility can never have too much data.
In Yorkshire Water’s case, the utility utilised new technology within OSIsoft’s PI system, a platform to manage complex industrial sensor data, and increased its loggers to 6,800 flow and pressure sensors as part of its ‘Visible Networks’ project.
Sewell says: “The goal of the project is to increase analytical capability using existing network data to get early event warnings, reduce nuisance alarms and see events that we’ve been able to visualise before.”
The early leak identification solution comprises Profile Alarms, threshold alerts based around personalised parameters to more quickly detect problems.
The automated monitoring of flows and pressures in the clean water distribution system allow the system to check for anomalies such as a sudden drop in pressure. An increase in flow at a meter, or chain of meters, may indicate a leak will bring such events to the attention of the relevant staff.
The project is at deployment stage and is being tested on live data with a 96% accuracy rate.
By using predictive analytics, says Sewell, Yorkshire Water can detect and resolve a leak within 10 days compared to 30 days to dig and repair a leak.
Harness data to predict rainfall
Another presentation at the OSIsoft conference demonstrated again the business value of data-based prediction.
United Utilities, a water distribution company in North-West England, presented how it is harnessing the power of predictive analytics to prevent combined sewer overflows.
By monitoring chamber water levels and responding to alarms to remove blockages or load balancing via an active system control, the utility can prevent the combined sewer overflows releasing untreated sewage into the environment during times of heavy rainfall.
United Utilities teamed up with the University of Exeter to develop a methodology based on the analysis of combined sewer overflows water levels as well as observed and forecasted rainfall data by means of Evolutionary Artificial Neural Networks.
The pair used PI System components to create a fully automated prototype system that can predict water levels up to six hours ahead with an 80-90% accuracy.
For Kevin Woodward, OT Strategy Manager at United Utilities, also presenting at the OSIsoft Users Conference, the business value of predictive overflow management is the potential to reduce Totex (Capex and Opex combined) as well as achieving regulatory and customer commitments.
Woodward gives an example of using real-time data to cut operational costs. United Utilities and the University of Exeter have developed algorithms to predict consumer water demand 24 hours in advance with 98% accuracy with the outcome of minimising pump usage and therefore energy consumption.
Dr. Michele Romano, who works for United Utilities in the Operational Technology data analytics team, asserts that the proof of concept shows that: “Agile development can be achieved by making real-serious data readily available. That allowed us to develop, test and validate methodologies.”
The next step for United Utilities is to scale up the project. The proof of concept was tested on two combined sewer overflows but there are plans to extend it to half a dozen more. “One of the main strengths of a data-driven neural network is that it is portable and can be moved across sites,” says Woodward.