Researchers at Exeter University are using computer game technology and artificial intelligence to help predict floods and spot leaking pipes.
A team from the university’s Centre for Water Systems is developing software that is better at detecting problems in a water distribution network by learning how to identify false alarms.
It is also using powerful computer processors designed for supporting complex 3D graphics to create models that can predict how rainfall or building work might affect an area’s probability of flooding, much more quickly than existing software.
‘It is important for a water company to get information [about leaking pipes] as soon as possible because then they can intervene before the loss of pressure is felt by the customer,’ Prof Dragan Savic, director of the Centre for Water Systems told The Engineer.
Water companies use thousands of pressure and flow sensors in their distribution networks, sending information to a control centre every 30 minutes to enable software to identify anomalies in the water flow, of which there are hundreds every day.
But around 80 per cent of these anomalies are false alarms or ’ghosts’ caused by problems with the sensor, the signal transmission or flooding due to heavy rain.
It is usually up to human operators to identify which anomalies are due to real leaks or to teach the software how to recognise a ghost. Exeter’s software, however, learns how to do this without human intervention within an hour of the leak occurring.
‘We predict what the real pressure and flow should be at certain hours of the day and when the measured signal departs enough from our prediction then we classify it as a leak,’ said Savic, ‘with time, if that departure increases, our certainty about it not being a ghost is higher.’
To predict flood patterns, the researchers are using another form of artificial intelligence called a cellular automaton, which is designed to model things such as the population growth of predators and prey in an environment or the spread of disease.
This enables them to provide real-time warnings of when heavy rainfall might cause a flood, risk analysis of a specific geographical area over time, or predictions of how new buildings or sewage systems might have an impact upon the likelihood of flooding.
In order to run the large number of calculations in the model at a useful speed, the researchers had to change the way the equations were treated so they could run them through a graphics processing unit (GPU) normally used for rendering video games.
The three-year research project is funded by the Engineering and Physical Sciences Research Council and 12 industrial partners. The team intends to make the finished programs available on an open-source licence so that any organisation or company can use them for free to develop their own software.