Water/energy consumption in water supply will be optimised and the resilience of resource management using advanced water demand forecast (Demo Case #2) will be improved. The demand management tool will be focused on predicting near-future demand behaviour and market trend. For that purpose, forecasting techniques (e.g. ARIMA, regression models) will be combined with classification techniques (e.g. neural networks, gradient boosting, X-means, kNN) to avoid inefficient water demand states and with multi-objective optimization techniques (e.g. Mixed Integer Linear Programming, MOGA, Particle Swarm Optimisation) to balance supply with demand. These methodologies are based on previous experiences like the WatERP project, and will be verified using EPANET simulations.
The detection, location and mitigation of water leakage in water supply networks will be improved (Demo Case #2 and #4). The leakage detection and mitigation tool is a recommender tool aimed at detecting, locating and proposing mitigation actions to leakage. This tool uses multi-hierarchy models combined with semantic reasoning and EPANET- p simulations to find main leak cause, derive the impact of the leak and propose remediation actions. Regarding bursts and leakage alerts, the anomaly detection algorithms will be based on Evolutionary Polynomial Regression (EPR) and/ or Bayesian networks, developed in previous projects (UNEXE, EUT). The implementation of resulting maintenance actions will be supported by a predictive maintenance tool (Demo Case #2) providing long-term forecasts of system malfunctions, and a workforce tool to optimally assign and schedule operator’s tasks according to maintenance operations to be performed. The predictive maintenance tool will use Bayesian networks as a knowledge base model to analyse the occurrence probability of maintenance events and the corresponding constraints. Supporting the Bayesian network model, forecasting tools (time series analysis), classification techniques (ANN), and clustering techniques (K-means), will be used to predict decalibrations in sensors, classify and detect failures in operations and calculate the dispersion or occurrence in similar functional systems. The workforce tool will be utilise multi-objective optimization algorithms in order to provide an optimal schedule for operator’s tasks based on the type of operation, the personnel expertise, the deadline (or SLA), the distance, etc. EUT will lead these efforts thanks to the expertise in ICT4Water solutions combined with experience and technology transfer from energy and industry domains like LowUP in the case of bringing Fiware4Water workforce management and predictive maintenance solutions.
Furthermore, early detection and reaction to water contamination events in water supply networks will be achieved here (Demo Case #2). Work will develop analytical services aimed at ensuring water quality in supply networks and detect contaminants in the water. An anomaly detection tool for water quality will be developed by TZW. Anomalies can be caused by a change in water quality, but also by faulty sensors. Especially in drinking water supply systems, where a large number of sensors are used. The problem of faulty sensors (caused by fouling, lack of calibration) is a source of false alarms. Therefore, the tool will be able to detect both types of anomalies. Technically, the tool will be based on higher order tensor models where the sensor network within a supply system is described by terms of several water quality parameters, locations and time. Together with soft-sensors (digital twin of real sensor) it will differentiate between true water quality changes and faults.
