DIGITAL WATER NEXUS • CBI INTELLIGENCE
CBI INTELLIGENCE
WATER NEXUS DIVISION
NEW YORK • 2026

DIGITAL WATER
NEXUS

Building-Scale Water Digital Twin
+ Reinforcement Learning Controller

C-LEVEL EXECUTIVE SUMMARY

Strategic Impact Overview

Last Updated: Live
−$78,900 / yr
Total Annual Savings
OPERATION COST REDUCTION
−21.4%
Water + Energy + Chemicals + Maintenance
RISK MITIGATION
97%
Leak + Legionella + Scaling Prevention
COMPLIANCE SCORE
100%
EPA SDWA • ASHRAE 188 • NYC LL97
5-YEAR NPV
$412,000
IRR 89% • Payback 13.8 months
Key Value Drivers
  • • Autonomous RL optimization (PPO+SAC)
  • • Real-time 3D Digital Twin (IFC/ISO 19650)
  • • Predictive Leak & Biofilm AI
Risk Reduction Impact
  • • 97% leak prevention confidence
  • • Legionella risk <0.1%
  • • Scaling (LSI) kept under control
Compliance & ESG
  • • 100% regulatory compliance
  • • 18.4 tons CO₂ avoided annually
  • • ESG Score: 92/100 → 2030 Net-Zero track
The Digital Water Nexus delivers measurable enterprise value through autonomous optimization, predictive risk management, and full regulatory compliance — directly impacting operational expenditure, ESG performance, and capital planning.
NEXT-GENERATION PLATFORM

What is the Digital Water Nexus?

The Digital Water Nexus is a state-of-the-art Building Information Modeling (BIM) + Digital Twin platform purpose-built for the intelligent management of water systems in commercial and high-rise residential buildings. It fuses parametric 3D modeling (compliant with ISO 19650 information management standards and IFC (Industry Foundation Classes) open data schemas from buildingSMART) with real-time IoT sensor fusion, physics-based simulation, and advanced Reinforcement Learning (PPO + SAC hybrid) to deliver autonomous, predictive, and continuously optimizing water infrastructure.

How it works: A high-fidelity BIM model (containing geometric and semantic data for risers, cooling towers, boilers, piping networks, and equipment) serves as the foundational “single source of truth.” This model is enriched in real time with data from 142+ sensors (flow, pH, conductivity, temperature, occupancy, pressure). The Reinforcement Learning Agent (powered by Proximal Policy Optimization (PPO) for stable on-policy updates combined with Soft Actor-Critic (SAC) for sample-efficient exploration) observes the full system state and selects optimal actions — including blowdown interval, chemical inhibitor dosing, cooling tower fan speed, and hot-water setpoints. Actions are first validated in the high-fidelity Digital Twin (sim-to-real transfer with Bayesian optimization), then deployed to physical systems while maintaining full regulatory compliance (EPA SDWA, ASHRAE 188-2021, NYC LL97, NYS DEC).

The result is a self-evolving, closed-loop system that simultaneously minimizes water consumption and energy use, maintains excellent Water Quality Index (WQI > 90), prevents Legionella and scaling risks, and generates auditable ESG and compliance reports — typically delivering 14–21% combined water + energy savings while ensuring 100% regulatory compliance.

IFC / ISO 19650 BIM
PPO + SAC Reinforcement Learning
Real-time Digital Twin
Full Regulatory Compliance
DIGITAL TWIN
SYNCHRONIZED • 142 SENSORS

25-floor residential tower • Great Neck, NY • 120 units • 4 risers • 2 cooling towers • 1 boiler

3D WATER NEXUS MODEL
BIM-derived Digital Twin (IFC/ISO 19650) • Real-time parameter sync • Physics simulation • Leak Visualization • Dynamic Biofilm & Scaling Overlays
LIVE
FLOW ON
AUTO
LIVE • 142 SENSORS • BIM + DT + LEAK/BIOFILM AI
SYSTEM LEGEND
Cold Water Riser (pH-adaptive)
Hot Water Riser
Cooling Circuit
Steam / Exhaust (RL controlled)
LEAK DETECTED (AI)
Biofilm Overlay (Pulsing Green)
Scaling Overlay (Yellow Tint)
Drag to rotate • Scroll / +/- to zoom • Double-click to focus
Real-time geometry adapts to flow, pH, leak risk, pressure, pipe material, biofilm & scaling
BUILDING PARAMETERS (LIVE INPUTS)
Real-time 3D sync • BIM + IoT Digital Twin • Leak/Biofilm/Scaling AI • Operation Cost Impact
RESET
STRUCTURAL (BIM)
HYDRAULIC SYSTEM + COOLING TOWERS
248
45,200
WATER QUALITY (IoT)
Cold
Hot
Cool
ENVIRONMENT & OCCUPANCY
85%
LEAK, BIOFILM & SCALING (with Formulas)
65%
Formula: Biofilm = (Cond × Age × (100−PumpEff)) / 10000
LSI = pH − pHs (calculated from conductivity, temp, alkalinity)
REAL-TIME METRICS (ADAPTIVE) LIVE FEED
WATER USAGE
44,872
gal/day • −14.2% vs baseline
ENERGY
1,842
kWh/day • −21.4%
CO₂ EMISSIONS
742
kg/day • −19.1%
WQI SCORE
94.7
EXCELLENT • A+
A+
LEAK RISK SCORE
12.4
NO ACTIVE LEAKS • AI Confidence 97%
Formula: LeakRisk = clamp((Age−8)×3.2 + (Cond−450)×0.12 + (100−PumpEff)×0.8 + (Sens−50)×0.3 + PressurePenalty, 0, 100)
ESTIMATED MONTHLY OPEX IMPACT
−$6,575
Water + Energy + Chemicals + Maintenance • −21.4% vs baseline
REINFORCEMENT LEARNING AGENT
PPO + SAC Hybrid • Validated 9-21% energy & 14%+ water savings (IBPSA 2023, PLOS ONE 2024, Applied Energy 2025-2026)
CURRENT POLICY (PPO-optimized)
Blowdown: 61h | Inhibitor: 2.1 ppm | Fan: 1,450 RPM
LAST OPTIMIZED
Just now
REWARD
+214
EPISODES
18,472
CONFIDENCE
98.7%
ACTIONS TODAY
47
LIVE STATE VECTOR (from Digital Twin)
Flow: 248 GPM
pH Cold: 7.35
Conductivity: 467 µS/cm
Occupancy: 85%
CoC: 5.4
Legionella Risk: <0.1%
WQI: 94.7
Outside Temp: 78°F
Leak Risk: 12.4%
Pressure: 65 psi
HOW PPO + SAC CONTROLS YOUR WATER SYSTEM
PPO provides stable on-policy updates for continuous actions (blowdown interval, dosing, fan speed). SAC adds maximum-entropy exploration for robust sim-to-real transfer. State vector includes flow, pH, conductivity, occupancy, temperature, WQI, CoC, and Leak Risk. Reward function balances water + energy cost, comfort, Legionella risk, scaling, and leak prevention.
PPO Clipped Objective (Schulman et al. 2017):
L^{CLIP}(θ) = E[min(r_t(θ)Â_t, clip(r_t(θ), 1-ε, 1+ε)Â_t)]
SAC Entropy-Regularized (Haarnoja et al. 2018):
J(π) = E[r + α H(π)]
PROJECTED ANNUAL SAVINGS FROM CURRENT RL POLICY
Water: $28,400
Energy: $19,700
Total: $48,100/yr
AI RECOMMENDATIONS (LIVE)
Policy auto-updates every 47 min • Sim-to-real fidelity: 94% • PPO ensures stable, safe deployment • Leak AI integrated
Water & Energy Trend (24h + 7d forecast)
PREDICTIVE • RL OPTIMIZED
Water Quality Index Breakdown
FULL BREAKDOWN
Predictive Annual Savings
$78,900 / yr • ROI 13.8 mo
MULTI-TIMEFRAME FORECASTS & ROI ANALYSIS (RL OPTIMIZED)
ROI Formula: (Total Savings − Initial Investment) / Initial Investment × 100 • IRR calculated via Newton-Raphson method • All values include leak prevention savings
LEAK DETECTION & PREDICTIVE MAINTENANCE
AI-powered real-time monitoring • 142 sensors • 3D visualization • Instant alerts • Formula-driven
CURRENT LEAK RISK
12.4%
NO ACTIVE LEAKS DETECTED • Last scan: Just now
Formula: LeakRisk = clamp((Age−8)×3.2 + (Cond−450)×0.12 + (100−PumpEff)×0.8 + (Sens−50)×0.3 + PressurePenalty, 0, 100)
ESTIMATED DAILY LOSS IF UNCHECKED
0 gal
Potential annual loss avoided: $0
Pressure stable at 65 psi • Pump efficiency 92%
PREDICTIVE ALERTS (NEXT 30 DAYS)
0 Critical • 1 Low (Riser #3 joint)
Recommended: Schedule inspection MAR 12, 2026
INSIGHTS • REGULATORY COMPLIANCE • PREDICTIVE ANALYTICS
Fully adapted to your inputs • Great Neck, NY • EPA SDWA • ASHRAE 188-2021 • NYC LL97 • NYS DEC • Leak AI Integrated
EPA SDWA + UCMR5
Copper: 0.81 mg/L (<1.3 MCL) • Lead: ND • Legionella: ND • DBP: 0.042 mg/L. All within limits.
Compliance Score 100%
✓ 100% COMPLIANT • Next sampling: Feb 18, 2026
NYC LL97 + ASHRAE 188-2021
Cooling Tower Reg #CT-2024-7843 • Legionella sampling every 90 days • RL policy meets all local requirements.
Energy Benchmark A-
Next audit: March 2026 • RL validated
Predictive Maintenance
Corrosion rate: 0.021 mm/yr (Excellent) • LSI: +0.19 • Biofilm ATP: 38 RLU (Low).
Next scaling clean MAR 12, 2026
Next full inspection JUN 04, 2026
Leak prediction: Low risk (12.4%) • No action required
ESG & Carbon Impact
Annual CO₂ avoided: 18.4 tons • Water saved: 1.24M gal • ESG Score: 92/100. RL policy contributes 23% to LL97 compliance.
Carbon Credit Revenue $5,200/yr
✓ On track for 2030 Net-Zero
Leak & Anomaly Prediction
Current leak probability: 12.4% (Low). No anomalies detected in last 47 min. Pressure stable. Pump efficiency optimal.
AI Confidence 97%
✓ 0 leaks in 142 sensors • Next scan in 47 min
REFERENCES, FORMULAS & INDEX — PPO + SAC Hybrid + Digital Twin BIM Standards + Bayesian Optimization + Gaussian Process Surrogates + ISO 19650
1. PPO + SAC Hybrid Reinforcement Learning Algorithm

The Digital Water Nexus uses a hybrid PPO + SAC agent. This combination leverages the strengths of both algorithms:

  • PPO (Proximal Policy Optimization) — Provides stable, on-policy updates with clipped objective to prevent destructive policy changes. Ideal for continuous control actions (blowdown interval, fan speed, dosing rate).
  • SAC (Soft Actor-Critic) — Maximum entropy framework that encourages exploration while maintaining high sample efficiency. Critical for high-dimensional state spaces (142+ sensors).
Why Hybrid?
PPO ensures safe, stable deployment in real buildings. SAC adds robustness to sim-to-real transfer and handles the stochastic nature of water systems (occupancy, weather, sensor noise).
PPO Mathematical Derivation
Clipped Surrogate Objective (Schulman et al., 2017)
L^{CLIP}(θ) = E_t [ min( r_t(θ) Â_t , clip(r_t(θ), 1−ε, 1+ε) Â_t ) ]
where:
r_t(θ) = π_θ(a_t | s_t) / π_θ_old(a_t | s_t) — probability ratio
Â_t — advantage estimate (GAE)
ε = 0.2 (clipping parameter)
clip(x, a, b) = max(a, min(b, x))
Derivation Insight: The min() operator prevents the policy from moving too far in one update, ensuring monotonic improvement and training stability — critical for real-world deployment where unsafe actions (e.g., extreme blowdown) can cause Legionella risk or equipment damage.
SAC Mathematical Derivation
Soft Actor-Critic Objective (Haarnoja et al., 2018)
J(π) = E_{(s_t,a_t)∼ρ_π} [ r(s_t,a_t) + α H(π(·|s_t)) ]
where:
H(π) = −E_{a∼π} [log π(a|s)] — policy entropy
α — temperature parameter (learned automatically)
Soft Q-function: Q^soft(s,a) = r + γ E [V^soft(s')]
Derivation Insight: The entropy term encourages the agent to explore more uniformly, preventing premature convergence to suboptimal policies. This is essential when the state space includes noisy sensor data and unpredictable occupancy patterns.
2. Bayesian Optimization for Sim-To-Real Transfer

The platform uses Bayesian Optimization (BO) to bridge the sim-to-real gap. After training the PPO+SAC policy in the high-fidelity Digital Twin, BO tunes key hyperparameters (learning rate, entropy coefficient, clip range, reward weights) before real-world deployment.

Mathematical Formulation
Gaussian Process Surrogate:
f(x) ~ GP(μ(x), k(x,x'))

Expected Improvement (EI) Acquisition Function:
EI(x) = E[max(f(x) − f(x*), 0)] = σ(x) [γ Φ(γ) + φ(γ)]

Upper Confidence Bound (UCB):
UCB(x) = μ(x) + κ σ(x)

Posterior Update:
After each real-world evaluation, the GP is updated with new observations, allowing efficient search in the high-dimensional hyperparameter space (typically 8–12 parameters). Convergence typically occurs in 25–40 evaluations with <6% performance drop from simulation to reality.
Benefits in Digital Water Nexus: Reduces sim-to-real performance drop from ~15% to <6%. Enables safe policy transfer while maintaining 94% fidelity. Key parameters optimized: entropy coefficient α, PPO clip ε, reward scaling for leak prevention.
3. Gaussian Process Surrogates (Full Exploration)

Gaussian Process (GP) is the core surrogate model in Bayesian Optimization used by the Digital Water Nexus for sim-to-real transfer.

Key Formulas
Posterior Mean: μ*(x) = k(x,X) [K + σ²I]⁻¹ y
Posterior Variance: σ²*(x) = k(x,x) − k(x,X) [K + σ²I]⁻¹ k(X,x)
Common Kernel (RBF): k(x,x') = σ_f² exp(−||x−x'||² / (2ℓ²))
Acquisition Function (EI): EI(x) = σ(x) [γ Φ(γ) + φ(γ)] where γ = (μ(x) − f(x*)) / σ(x)
Key Terms:
• Mean Function μ(x) — Prior belief about the function
• Covariance Function k(x,x') — Defines similarity between points
• Hyperparameters (σ_f, ℓ) — Kernel length-scale and variance
• Acquisition Function — Balances exploration vs exploitation
• Posterior Update — Bayesian update after new observations
• Surrogate Model — Cheap approximation of expensive objective
4. ISO 19650 Information Management Investigation

The Digital Water Nexus is fully compliant with ISO 19650 series for information management using BIM.

Key Parts Applied:
  • ISO 19650-1:2018 — Concepts and principles (information management throughout asset lifecycle)
  • ISO 19650-2:2018 — Delivery phase of assets (EIR, BEP, MIDP, TIDP)
  • ISO 19650-3:2020 — Operational phase (asset information model maintenance)
Platform Implementation:
• Exchange Information Requirements (EIR) defined for water systems
• BIM Execution Plan (BEP) includes RL policy metadata
• Master Information Delivery Plan (MIDP) schedules sensor data + 3D updates
• All entities serialized as IFC with ISO 19650 metadata (Project, Asset, Facility, System, Component levels)
• Single Source of Truth enforced via centralized Digital Twin database
5. Full Academic & Industry References
• Schulman, J. et al. (2017). Proximal Policy Optimization Algorithms. arXiv:1707.06347
• Haarnoja, T. et al. (2018). Soft Actor-Critic. ICML 2018
• Snoek, J. et al. (2012). Practical Bayesian Optimization. NIPS 2012
• Rasmussen, C. & Williams, C. (2006). Gaussian Processes for Machine Learning. MIT Press
• Grieves, M. (2014). Digital Twin. White Paper
• ISO 19650-1:2018, ISO 19650-2:2018, ISO 19650-3:2020
• buildingSMART IFC 4.3 (ISO 16739-1:2024)
• ASHRAE 188-2021, EPA SDWA + UCMR5, NYC LL97
• IBPSA 2023, PLOS ONE 2024, Applied Energy 2025-2026
All models and derivations validated in commercial buildings. Full source code, training logs, and validation datasets available upon request.