Software modelling predicts water consumption and helps plan future investments

Water services provider Veolia has used Australian software in a European trial of water demand prediction, achieving an accuracy rate of 99.5%.

Veolia wanted to create a detailed simulation to model more than 160,000 residential water consumers in Nice, France.

Veolia Environment Research & Innovation (VERI) Urban Modelling expert Dr David Mouquet, said: “We were looking for a mathematical alternative to classical statistical solutions (linear regression, time series analysis, etc) that would allow us to draw some scenario of long-term evolution of water demand of the residential sector.”

The company used the Simulait software, an online tool for simulation-based consumer analytics, which was developed by Australian company Intelligent Software Development (ISD).

The software simulates a population of consumers to help predict how people will behave and allows users to run scenarios to test options for influencing that behaviour.

“We chose Nice city for two reasons. First, because water consumption there is mainly residential (95%) and Simulait had already been used for residential water demand. Secondly, because our business unit had already collected a lot of water consumption data on Nice,” Mouquet said.

Veolia used historical water demand data from the year 2000 to calibrate Simulati’s model. The company then forecasted data from 2001 through to 2010.

“We used a modelling technique which is quite different to classical statistical methods. Simulait is based on an agent-based modelling approach. We focused on the human behaviour related to water consumption, including how much households consume during their quotidian activity according to their sociodemographic characteristics (household composition, ageing, incomes, …), and their dwelling  characteristics (house or apartment).”

The organisation also looked at external influences on human behaviour, like meteorological conditions and water restrictions imposed by local authorities.

“For our case we used both data already implemented in Simulait (mainly water consumption ratios of appliances, human behaviour rules and appliance technology penetration in residential sector) and data specific to Nice city (census data, garden composition, meteorological data and, of course, water consumption).”

Some preliminary work was necessary, to geographically aggregate the data, Mouquet said.

The company then matched the predictions from Simulait to the real-world 2011 usage figures - the most recent year for which data was available - and found the model obtained an accuracy level of 99.5%.

“Thanks to historical data, we were able to validate the model developed in the context of Nice city.”

The company has used the software to look beyond 2011 into the future.

“Thanks to the modelling, we were able to see possible trajectories of water demand up to 2020. This information is important for our business unit, as it allows us to discuss water prices with local authorities and also to plan investments related to water production equipment.”