Welsh-Water - Service Reservoir Bacterial Compliance Predictive Model
Purpose of the Project:
To better understand the factors that lead to a service reservoir bacterial non-compliance, and to usen this information to build a statistical predictive model that can help predict where non-compliances are likely to occur.
Project Objectives as linked to our 25 Year Vision:
To ensure that customers have complete confidence that their drinking water is safe, reliable and tastes good, through:
- Environmental and asset monitoring
- New innovative monitoring systems which will allow us to continuously monitor key water quality parameters
Welsh Water owns 350 service reservoirs which supply over 800 million litres of safe, clean drinking water every day. A water sample is taken at each service reservoir each week and is tested at our laboratory for the presence of certain bacteria — specifically total coliforms and E. Coli. The sampling programme, regulated by the Drinking Water Inspectorate (DWI), is in place to ensure we comply with Water Quality regulations, and continue to provide high quality water to our customers. Around 18,000 regulatory water samples are taken at our service reservoirs every year. The vast majority of the samples we take pass our tests.
We very rarely find the presence of bacteria but when we do, we always investigate but this can be costly and can disrupt our usual operational processes. In 2015, one of our Statisticians completing an MSc at Cardiff University, through Welsh Water’s sponsorship programme, began a project to develop a statistical predictive model for bacterial non-compliances at our service reservoirs.
The project drew upon knowledge and expertise from within the company, and from the wider industry, to identify the possible factors that may lead to a bacterial non-compliance. Statistical techniques were then used to test the statistical significance of the suggested factors. Relevant data covering the last six years was explored to identify trends and patterns.
The improved understanding of contributory factors, unearthed using these statistical processes, was finally used to construct a statistical predictive model that highlights which service reservoirs have a higher probability of incurring a bacterial non-compliance; this can help predict where failures may occur in the future.
Once verified, this information can be used to help inform investment decisions and enable us to reduce the risk of incurring bacterial non-compliances at our service reservoirs. An initial model, produced earlier in the year, was able to reliably predict around 70% of noncompliance events. Further work on this project, which will involve incorporating further contributory factors into the model, should increase this accuracy even higher.
For more information about this
project contact Kevin Parry, Statistician