Preventing rising main pollutions through machine learning

Mounting pressure on utilities to achieve zero pollution events has accelerated sewer investment and action planning but rising main sewers pose a unique challenge. Better analysis of existing data can mitigate risks, says George Heywood , analytics innovation lead for technology specialist Ovarro.

Water companies in the UK are working on pollution reduction strategies with a sense of urgency, in line with Environment Agencyand Ofwat expectations of more action.

Requiring particular focus will be rising main sewers – pressurised pipes that pump wastewater from a pumping station towards a treatment works. These are high risk, critical assets but with many in the UK ageing and becoming more vulnerable to bursts, historic programmes of proactive maintenance and investment may no longer be enough to keep up with the rate of deterioration.

Often situated in hard-to-reach, remote locations, including beneath rivers, railways and roads, and in environmentally sensitive areas, a burst rising main can have catastrophic ecological impact, with consequences that are unacceptable in the eyes of customers, regulators and stakeholders.

Technical and logistical limitations in rising main monitoring can mean utilities are alerted – often by a member of the public - hours or even days into the event. This is too late to take action that would prevent a pollution, so the likely result is a costly clean-up operation, financial penalties, prosecutions, and long-term reputational damage. 

Advances in data science and artificial intelligence, including machine learning, mean the sector is now able to go beyond the basics of, for example, setting alarm thresholds on high flow rates. Sophisticated analysis of readily available data in near real-time means that much more reliable monitoring of rising main activity is now possible.

BurstDetect, Ovarro’s cloud-based early-warning system, accepts data at a range of monitoring frequencies with an algorithm being applied to understand and characterise ‘normal’ pumping station behaviour. . If a potential burst is detected an alert is sent to control rooms, often within 30 minutes of the incident occurring, significantly improving historic reaction time.

This “training and testing” approach to machine learning is becoming increasingly important to water companies, giving them more actionable insight than ever before, utilising data that may not have been fully harnessed otherwise.  With so much available data, it is just not possible for humans to process and analyse the information themselves. By having the correct technology and processes in place, the stage will be set for utilities to rapidly increase their real-time and predictive capabilities.

Automated algorithms, such as those created for BurstDetect, can always be improved and as water companies begin to implement the technology, Ovarro’s data scientists will work with them to assess the accuracy of alerts. By growing this dataset, through continuous feedback, BurstDetect’s algorithms can learn and the data science teams can improve the technology continually.

Utilities have thousands of pumping stations to monitor but rising main pollution is considered inexcusable by the regulators, with the wider community becoming increasingly sensitive to incidents that impact the environment. Having access to the latest digital technologies puts pollution prevention in the hands of operators and will play a significant role as utilities seek to demonstrate and deliver their commitments to environmental stewardship.

For more information: Ovarro.com/BurstDetect

Taxonomy

• Water
• Public Health
• Technology
• Environment
• Infrastructure