Goal

• Active leakage control to
  • increase the efficiency of water resources management in urban water networks.
  • Reduce waste of energy and water.
• Optimal control of pumps to reduce energy costs:
  • Demand forecast driven optimization.
  • Online learning and optimization (reinforcement learning).

Deployed Services Description

• Leakage localization:
  • Simulation of several leakage scenarios for the computation of induced flow and pressure variations.
  • Machine Learning for inverting the relation: inferring the set of (simulated) scenarios associated with the actual flow and pressure data (from sensors).
• Demand forecasting:
  • Time series clustering for the identification of typical patterns.
  • Learning a forecasting model for each identified pattern.

• Pump scheduling optimization:
  • Global Optimization using hydraulic simulation and demand forecasts.
  • Reinforcement Learning for online control/optimization.

Results

• Accurate (water)demand forecast (MAPE -Mean Avg Percentage Error lower than 2-3%) and anomaly detection (on smart metering data).
• Leakage localization(error reduction up to 1/5).
• Pump scheduling optimization(5-10% costs reduction).