01 The System, Not the Pump
The core shift in municipal pump station optimization is moving from "which pump meets the design point" to "which configuration minimizes total cost of ownership while reliably serving variable demand over 20+ years." The pump itself accounts for only 5–15% of lifecycle cost; electricity dominates at 80–90%.
Core principle
A 1% efficiency gain, in net present value terms, is usually worth more than the entire purchase price of the pump. Engineering effort belongs in control logic, correct sizing, and lifecycle analysis — not in the hardware alone.
02 Pump Sizing Math
Centrifugal pump selection is the intersection of the pump's Q–H curve and the system resistance curve, governed by four variables: flow \(Q\), total dynamic head \(TDH\), shaft power \(BHP\), and net positive suction head \(NPSH\).
Flow velocity
\[ v = \frac{Q}{A} = \frac{4Q}{\pi D^2} \]
Total Dynamic Head
\[ TDH = h_s + h_f + h_p + h_v \]
| Term | Meaning |
|---|
| \(h_s\) | Static head — elevation difference between discharge and suction |
| \(h_p\) | Pressure head — pressure difference between vessels |
| \(h_v\) | Velocity head, \(v^2/2g\) |
| \(h_f\) | Friction loss (Darcy–Weisbach or Hazen–Williams, below) |
Friction loss — two methods
Darcy–Weisbach: \( h_f = f \cdot \dfrac{L}{D} \cdot \dfrac{v^2}{2g} \)
Hazen–Williams: \( h_f = \dfrac{10.67 \cdot L \cdot Q^{1.852}}{C^{1.852} \cdot D^{4.87}} \) (SI units; \(C=130\text{–}150\) for new plastic pipe, lower for corroded iron)
Power and efficiency
\[ P_{hyd} = \rho \cdot g \cdot Q \cdot TDH \qquad BHP = \frac{P_{hyd}}{\eta_{pump}} \]
Example: 45 hp hydraulic load at \(\eta = 0.78\) → BHP = 57.7 hp → select a 60 or 75 hp motor.
Cavitation check (NPSH)
\[ NPSH_A > NPSH_R + \text{margin (0.6–1.0 m)} \]
Re-check at every VFD speed point — NPSHR shifts with speed.
03 Lifecycle Cost & ROI
Compare options on net present value over the full service life, not upfront price. Unaccounted water — 15–30% lost to leaks in typical systems — is pumped and paid for, then lost outright.
Efficiency levers
- Size to actual demand. Stacked safety factors (peak × growth × fouling × service margin) push a pump permanently left of its best efficiency point (BEP). Size to the weighted average operating point instead.
- VFDs. Power scales with the cube of speed — dropping to 80% speed cuts consumption to ≈51%. Limit: high static-head systems have a minimum speed below which flow stops.
- Tariff scheduling. Shifting pumping to off-peak windows and trimming the 15-minute peak demand cuts demand charges without new hardware.
- IE3/IE4 motors. Efficient at 75–100% load; an oversized motor at 30% load is worse on both efficiency and power factor.
- Pressure management. Minimum sufficient pressure cuts leakage, extends pipe life, and saves energy directly.
- N+1 staging. Several smaller parallel pumps give redundancy while holding the operating point near BEP across a wide flow range.
Worked example — 100 kW station, continuous duty, €0.12/kWh
| Metric | Before | After |
|---|
| Annual consumption (kWh) | 876,000 | 744,600 |
| Annual energy cost (€) | 105,120 | 89,352 |
| Annual saving | €15,768 (15%) |
| Payback period | ≈ 1.14 years |
\[ NPV = \sum_{t=1}^{20} \frac{E_{saved}}{(1+r)^t} - C_{invest} = 15{,}768 \times 9.82 - 18{,}000 \approx €136{,}800 \]
8.6×
Base case · 15% saving · 88%/yr ROI
4.3×
Moderate · 10% saving · 50%/yr ROI
2.5×
Conservative · 8% saving · 25%/yr ROI
Return per €1 invested over the 20-year lifecycle. Even the conservative case outperforms most financial instruments.
04 Demand Variability
Demand swings 2–4× within a day, 10–30% between weekdays and weekends, and 30–80% between summer and winter. A pump sized only for peak sits far from BEP most of the time; one sized only for average can't cover peaks.
Daily pattern
- Morning peak (6:00–9:00): \(K_h = 1.65\text{–}1.85\); residential peaks are higher and narrower than commercial.
- Evening peak (18:00–20:00): comparable to or exceeding the morning peak in hot climates.
- Night minimum (1:00–5:00): \(K_h = 0.24\text{–}0.40\); pressure is highest and background leakage peaks — a persistently elevated night flow signals leaks.
- AWWA M22 design standard: \(K_h \approx 1.8\) typical day, \(K_h^{max} = 2.0\text{–}2.5\) for design.
Weekly and seasonal pattern
- Residential demand peaks Saturday/Sunday (1.2–1.4× average); commercial peaks on weekdays, dropping 20–40% on weekends.
- Summer peak month vs. average month: 1.3–1.8×; with lawn irrigation, up to 2.0×+ and hourly \(K_h > 10\) during watering.
- Tourist areas show inverted seasonality; industrial demand tracks production cycles.
| Level | Range | Coefficient |
|---|
| Peak hour / average day | 1.8–2.5 | \(K_h\) |
| Max day / average day | 1.5–2.5 | \(K_d\) |
| Max month / average month | 1.3–1.8 | \(K_m\) |
| Summer / winter | 1.3–2.0 (3.0+ with irrigation) | — |
05 Optimization Formulation
Generalized demand profile:
\[ Q(t,d,s) = Q_{avg} \cdot k_h(t) \cdot k_d(d) \cdot k_s(s) + Q_{fire} + Q_{leak} \]
Pump-scheduling objective over a 24–168 hour horizon:
\[ \min_{u(t)} \sum_t \Big[ \underbrace{c_e(t)P(t)\Delta t}_{\text{energy tariff}} + \underbrace{c_d P_{peak}}_{\text{demand charge}} + \underbrace{c_{start}N_{start}}_{\text{start wear}} + \underbrace{c_p\max(0,p(t)-p^*)^2}_{\text{pressure penalty}} \Big] \]
Subject to: minimum nodal pressure, tank levels within bounds, pump speed/cycling limits, \(NPSH_A > NPSH_R + \delta\) at every operating point, and N+1 redundancy.
06 Solution Approaches
- Demand profiling. Twelve+ months of hourly SCADA/AMI data, normalized by weekday, Saturday, Sunday/holiday, and season.
- Short-term forecasting. "Similar-day" method for limited data; ML (random forest, gradient boosting, SVR/MLP) where SCADA/AMI volume allows.
- Pump scheduling (EPANET). Hourly patterns for demand, head, pump speed, and tariff — typically 10–30% energy savings with no new hardware.
- Tanks as demand buffer. Fill overnight, draw down during peaks.
- Staged VFD pump fleet. Small pump for night/base flow, larger units layered in for daily and peak demand.
- Dynamic pressure control (SCADA/PRV). Automatic night-time pressure reduction (0.7–1.4 bar, 23:00–5:00), returning to day setpoints unattended.
- Minimum Night Flow (MNF) monitoring. A persistently elevated MNF flags leaks or bad pressure setpoints — without it, the optimizer will "optimize" pumping losses instead of eliminating them.
07 10-Step Field Checklist
- Map infrastructure: pipe lengths, elevations, fittings, bends.
- Build demand profile \(Q(t,d,s)\): weekday, weekend, holiday, season.
- Set design flows: \(Q_{avg}\), \(Q_{peak}\), fire scenario, N+1 failure.
- Build the system curve: Darcy–Weisbach losses across \([Q_{min}, Q_{max}]\).
- Select pumps: BEP at the most frequent flow, not just the peak.
- Size VFDs and parallel configurations to real demand profiles.
- Size BHP: motor with margin, without excessive oversizing.
- Verify \(NPSH_A\) at all design speeds, levels, temperatures, and VFD points.
- Optimize the schedule against tariffs, tank levels, and start count (Section 5).
- Monitor continuously: MNF, vibration, motor current, bearing temperature, actual efficiency.
08 The EPANET Network File (.inp)
Every calculation in this reference — and the working tool in Section 9 — depends on one input: an EPANET-format .inp file describing the actual network. EPANET (U.S. EPA) is the standard open hydraulic simulation engine for water distribution systems; WNTR (Sandia National Laboratories) wraps it in Python, which is what the pump-scheduling optimizer below is built on.
An .inp file is a plain-text description of network topology and behavior, organized into labeled sections:
| Section | Contents |
|---|
| [JUNCTIONS] | Demand nodes: elevation, base demand |
| [RESERVOIRS] | Fixed-head sources — a river intake, treatment plant outlet |
| [TANKS] | Storage: elevation, diameter, min/max/initial level |
| [PIPES] | Links: length, diameter, roughness coefficient |
| [PUMPS] | Head-flow curves and, optionally, speed patterns and efficiency curves |
| [PATTERNS] | Time-varying multipliers — demand, tariff, pump speed |
| [CURVES] | Head-flow and efficiency curve point data |
| [CONTROLS] / [RULES] | Status and operating logic — e.g. "pump ON if tank level below X" |
| [OPTIONS] / [TIMES] | Simulation duration, timestep, units, headloss formula |
Where to get one: most SCADA/GIS platforms can export directly to EPANET INP format; the free EPANET desktop application (or its Python wrapper, WNTR) can also build one from a hand-drawn network or convert from other formats. Without a calibrated model, a pump-scheduling optimization has nothing reliable to optimize against — the model quality sets the ceiling on the result quality.
09 This Project — A Working Beta
Everything in Sections 5–7 above — demand-aware scheduling, tariff-driven cost minimization, hydraulic feasibility constraints — is implemented as a working tool, not just a formulation. Upload an .inp file and a time-of-use tariff, and it runs a real extended-period EPANET simulation plus a differential-evolution search for a lower-cost pump schedule, checking tank levels, minimum service pressure, and demand at every step of the day.
Honestly: this is a beta. The optimizer and hydraulic model are genuine and independently tested — not a mockup — but the tool is new. Treat results as a starting point for engineering review, not an unattended operational decision.
Try the tool →
10 Conclusions
This analysis shows that optimizing municipal water supply is a whole-system management problem, not a matter of selecting an individual pump unit. Lifecycle cost is most strongly influenced by pump operating regimes, demand variability, the network's hydraulic characteristics, and equipment control strategy, while the initial cost of the pumps themselves accounts for only a small share of total expenditure.
Using real demand profiles, a mathematical model of the hydraulic system, and pump-scheduling optimization methods keeps equipment operating points close to the best efficiency point (BEP), reduces electricity consumption, lowers start counts, limits excess network pressure, and improves the reliability of water supply.
The most effective engineering measures are variable-frequency drives, staged operation of multiple pumps, storage tanks to smooth daily demand variation, dynamic pressure control, and continuous monitoring of network parameters. Applied together, these measures reduce operating costs, extend equipment service life, and make the system more resilient to load changes.
Operating experience and research results show that comprehensive pumping-station optimization can cut energy consumption by 10–30 % without degrading water supply quality, and in most cases delivers strong investment returns through reduced electricity and maintenance costs.
The engineering approach presented here can serve as a methodological foundation for designing new pumping stations, upgrading existing facilities, and developing automatic control algorithms for municipal water supply systems.
11 References
- Kossieris et al. (2018). Statistical properties of residential water demand at fine time scales. Water, 10(10), 1481.
- Młyńska, A. (2022). Hourly water consumption structure in selected households. J. Ecological Engineering, 23(9), 219–230.
- Bergel, Szeląg & Woyciechowska (2017). Seasonal influence on demand patterns, rural supply line. J. Water and Land Development, 34, 59–64.
- AWWA (2004). Manual M22: Sizing Water Service Lines and Meters (2nd ed.). Denver: AWWA.
- Di Mauro et al. (2020). Urban water consumption at multiple scales — dataset review. Water, 13(1), 36.
- Arampatzis et al. (2004). Decision support for water distribution management. Advances in Engineering Software, 35(2), 121–130.
- Georgescu & Georgescu (2015). Pumping station scheduling in EPANET. IOP Conf. Series: Earth Environ. Sci., 664.
- Salomons et al. (2020). Real-time optimization for energy-efficient water distribution. J. Cleaner Production, 275, 124112.
Condensed English edition of the original Russian-language reference document (26 May 2026). Re-derive local \(K_h\), \(K_d\), \(K_m\) coefficients from SCADA/AMI data at least annually.