Reducing alarm overload in water networks with AI and machine learning

Every day, thousands of alarms are triggered across water and wastewater networks, warning operators of potential problems from pressure changes to equipment faults. As regulatory monitoring requirements have expanded, so has the number of sensors feeding data into the system, but the team managing it has not grown to match.

Control room operators at Northumbrian Water manage around 45 alarms an hour each, peaking at 200 – well above the recommended 30 – and more than double what the same two-person team handled over 20 years ago. With so many alarms coming in, it is important to identify which ones need urgent action to minimise customer disruption.

We're developing Smart Alarm Management (SAM) using machine learning and data modelling to prioritise the most important alerts and group them by root cause. Generative AI is then used to gather and present relevant insights from other parts of the data and recommend what action to take. This means that an event that might once have appeared as 20 separate alarms can be presented as a single, consolidated item, with priority ranking, historical insights, and recommended actions.

By prioritising alerts and surfacing root cause, SAM is intended to help control teams get engineers to the right place faster, reducing the risk of interruptions to supply and providing better protection for the environment by helping prevent leaks or spills.

The proof-of-concept reduced alarm volume by 75% through clustering, confirming the approach on real Northumbrian Water data before committing to full development. That result underpinned a joint bid to Ofwat's Innovation Fund submitted with Northumbrian Water, Wessex Water, Anglian Water, and EEMUA, which secured £1.7m to develop and deploy SAM.