Water utilities have long depended on forecasts that are good enough most of the time. But in day-to-day operations, “most of the time” is not enough. If a crew takes part of a network offline for inspection or repair and demand turns out to be higher than expected, pressure can fall below acceptable levels, customers can lose supply and operators can face contractual penalties.
A study published in Neural Computing and Applications argues that the real breakthroug...
Continue Reading This Article
Enjoy this article as well as all of our content, including reports, news, tips and more.
By registering or signing into your SRM Today account, you agree to SRM Today's Terms of Use and consent to the processing of your personal information as described in our Privacy Policy.
h is not simply a more accurate prediction model, but a system that can judge when its own forecast is unreliable and adjust maintenance plans accordingly. The framework, called CAUCCES, was developed by Mohammadhossein Homaei and colleagues and tested across twelve Spanish municipalities over 18 months.
The researchers combined four forecasting methods: a dual-branch LSTM network, Prophet with weather inputs, and two gradient-boosting models, LightGBM and XGBoost. Together, the ensemble produced a mean absolute percentage error of 14.12%, ahead of benchmark systems such as DeepAR and the Temporal Fusion Transformer. The team said the gain came from model diversity rather than sheer complexity, with each component capturing a different pattern in water demand.
At the centre of the system is the Explainable Confidence Index, or ECI, a measure designed to show when the forecast should be trusted and when it should not. Rather than relying on costly Bayesian sampling, the index uses the spread between model outputs and the degree of disagreement among their weights. That makes it fast enough for live scheduling, while still giving operators a practical way to separate routine jobs from those that should be delayed.
The impact on operations was substantial. According to the study, SLA breaches fell from 9.3% to 1.5% when scheduling was driven by ECI, while operational costs rose by only 3.1%. The framework also cut completion times, emergency response times, customer disruption, fuel use and emissions, suggesting that better timing can improve both service reliability and efficiency.
The authors also stress that the system is not a universal fix. Performance weakened outside the development region in Spain, particularly when the model was moved into areas with different climate and consumption patterns. They said the framework needs local recalibration and a run of neighbourhood data before deployment, and that it is best suited to rural and smaller urban utilities rather than very large networks.
Even with those limits, the study points to a broader shift in how infrastructure software is being designed. Instead of asking only what demand will be, CAUCCES asks how confident the system is about that answer and whether maintenance should wait. For utilities working with thin margins and ageing assets, that distinction may matter as much as the forecast itself.
Source: Noah Wire Services
