Leveraging AI for Smart Water Network Optimization (Case Study: Georgia)
Access to safe, reliable water services is central to SDG 6: Clean Water and Sanitation, which calls for universal and equitable access to safe drinking water and improved water-use efficiency. Yet across many regions, especially small and mid-sized municipalities, water distribution systems suffer from pressure instability, leakage losses, energy inefficiency, and limited technical planning capacity. These inefficiencies not only waste water and energy but also threaten long-term sustainability.
In Khelvachauri, Georgia, a municipality near Batumi, researchers and engineers demonstrated how artificial intelligence (AI), machine learning, hydraulic modeling, and decision-support systems can modernize water network management. By integrating AI into traditional hydraulic simulations, the initiative delivered measurable improvements in system performance, offering a scalable, data-driven pathway to strengthen water infrastructure and directly advance SDG 6.
Context and Challenge
The potable water distribution system in Khelvachauri faced structural and operational challenges typical of aging infrastructure in resource-constrained contexts. Historical expansions were conducted without comprehensive hydraulic modeling, leading to: pressure deficits (especially in southern zones during peak demand), elevated non-revenue water (NRW) due to leakage, inefficient pumping operations, and high energy consumption.
These issues reduced service reliability and increased operational costs. Traditional hydraulic optimization requires specialized engineers and proprietary tools, often inaccessible to smaller utilities with limited financial and technical resources.
Without improved analytical capability, the municipality risked continued water losses, service inequities, and inefficient infrastructure investments, all directly undermining SDG 6 targets on water-use efficiency and sustainable water management.
Data-Driven Solution
To address these gaps, engineers enhanced the MACS Water Smart (WS) application, integrating:
- EPANET-based hydraulic simulation
- Geographic Information Systems (GIS)
- Random Forest machine learning models
- Automated optimization algorithms
The system was designed as a decision-support platform capable of replicating expert-level hydraulic optimization through AI.
Methodology
The process began with detailed network data collection, including: pipe diameters and lengths, elevation and topography, pressure and flow measurements, and network topology.
This data was calibrated within EPANET simulations to establish a baseline model of system performance.
A Random Forest regressor, a supervised machine learning algorithm, was then trained using simulation outputs and engineering rules. The model learned to predict optimal pipe diameter configurations and operational adjustments that would meet performance targets while respecting velocity and pressure constraints.
The AI effectively automated what would traditionally require extensive expert analysis, identifying infrastructure upgrades and system adjustments to improve performance.
Implementation and Results
The enhanced Water Smart system was applied to the Khelvachauri network model. The implementation followed three main stages:
Stage 1: Baseline Simulation: Initial modeling confirmed widespread pressure deficiencies and suboptimal flow velocities, particularly during high-demand periods.
Stage 2: AI-Based Optimization: The Random Forest model generated optimized pipe diameter recommendations and network configuration adjustments.
Stage 3: Re-Simulation and Validation:
After applying AI-generated modifications, the system was re-simulated.
The results were significant:
- Pressure increases exceeding 25 meters of water column (mwc) in critical low-pressure zones
- Flow velocities reduced below 1.5 m/s, enhancing hydraulic stability
- Pumping efficiency improved by 15%, reducing energy consumption
- Leakage rates reduced by approximately 8%, decreasing non-revenue water losses
- Computation time reduced by 35% compared to conventional optimization approaches
These measurable improvements demonstrate that the AI-enhanced system not only replicated expert engineering solutions but did so more efficiently and with reduced reliance on specialized human expertise.
Outcomes and Impact
- Improved Water Reliability
The increase of over 25 mwc in pressure levels restored stable service delivery, particularly in previously underserved southern areas. Residents experienced improved supply consistency, directly supporting equitable access to clean water.
- Enhanced Resource Efficiency
An 8% reduction in leakage translates into substantial water savings over time, while a 15% improvement in pumping efficiency lowers operational energy use, reducing both costs and environmental footprint.
- Democratization of Technical Expertise
By embedding engineering logic into machine learning models, the system reduces dependence on scarce hydraulic specialists. Small and mid-sized utilities can now access advanced optimization tools without large consulting budgets.
- Future Planning Capability
Because the platform integrates GIS and predictive analytics, it supports long-term infrastructure planning and scenario modeling, enabling utilities to prepare for population growth and changing demand patterns.
Conclusion
The Khelvachauri case illustrates how data, AI, and machine learning can directly advance SDG 6 by transforming traditional water management into an intelligent, evidence-based system. Through integration of hydraulic modeling and Random Forest algorithms, the initiative achieved:
- Pressure increases above 25 mwc
- Flow velocity stabilization below 1.5 m/s
- 15% improvement in pumping efficiency
- 8% reduction in leakage rates
- 35% faster optimization processing
These numbers represent tangible improvements in water reliability, efficiency, and sustainability for an entire municipality, and not just technological wins.
By lowering technical barriers and enhancing operational intelligence, AI-powered water management tools like Water Smart provide a scalable model for small and mid-sized utilities worldwide. In doing so, they demonstrate that data-driven innovation is not a luxury but a practical and impactful instrument for securing clean water and sustainable infrastructure in alignment with SDG 6.