Define the objective and constraints
Decide whether the search targets energy, cost, emissions, resilience or a combination. Explicitly enforce pressure, storage, pump operating envelopes and service requirements. A numerical optimum is only useful within a defensible feasible set.
Connect a search method to a model
An optimiser proposes a candidate schedule or asset configuration. A hydraulic solver evaluates the candidate. The result returns to the search process together with constraint violations. Evolutionary search and mathematical optimisation are different families; neither inherently guarantees a globally best, buildable solution.
Stress-test candidate solutions
- Re-run selected solutions in the full engineering model.
- Test different demand and boundary conditions, including asset outages.
- Inspect excessive pump switching and unrealistic control behaviour.
- Include end-of-period storage so that a cheap schedule cannot simply empty the tanks.
Report the trade-off
Present a small set of defensible alternatives with assumptions, sensitivities and operational implications. Do not reduce all environmental or cultural outcomes to an unexplained score. Strategic planning provides the decision context beyond the optimiser.
Formulate the feasible engineering problem
Decision variables might be pipe choices, pump schedules, valve settings or staged asset interventions. State their allowed values and physical implementation limits. A continuously variable pump-speed decision is inappropriate for a fixed-speed unit, and a schedule with dozens of starts may be unacceptable even if its energy calculation looks attractive.
Constraints should represent pressure, storage, source output, pump envelopes, switching limits and other adopted criteria. Define the assessment horizon and the scenarios under which the constraints must hold. A candidate that passes a single benign snapshot is not necessarily feasible through a full operating cycle.
Prevent an artificial energy saving
A common comparison error allows the optimised schedule to finish with less stored water than the baseline. The apparent energy saving then includes consumption of an initial reserve. Apply a justified terminal-storage requirement or a longer cyclic assessment so alternatives provide equivalent service and final state.
Use appropriate pump efficiency and the actual flow–head operating point. Electricity tariffs, carbon factors and energy boundaries should match the intended objective. Minimising tariff cost can shift energy use without reducing kWh, and minimising kWh does not automatically minimise lifecycle cost.
Read a trade-off rather than a single magic number
Suppose option A has a lower operating cost but less outage tolerance, while option B preserves more reserve at a higher cost. If neither dominates the other across the adopted objectives, the choice requires an explicit preference or constraint. A weighted objective hides that preference inside the weights unless it is reported clearly.
Keep costs, emissions, service and risk on defensible scales. Avoid large penalty values that merely disguise infeasibility without explaining which condition failed. Report constraint violations for rejected and shortlisted candidates so reviewers can understand what the search is trading away.
Use a solver loop with independent checks
- Generate a candidate within the permitted decision space.
- Run the engineering model with traceable inputs and consistent initial conditions.
- Record objective values, convergence status and every constraint violation.
- Search further according to the chosen method while retaining a reproducible run record.
- Re-run shortlisted candidates in the full model and inspect their operating behaviour independently.
A surrogate can accelerate screening, but it may be least accurate near the constraints that matter most. Verify promising candidates with the governing physical model. A failed hydraulic run is not a feasible low-cost solution and should not be assigned a misleading objective value.
Test robustness and practical delivery
Challenge shortlisted options with demand uncertainty, lower source levels, asset outages and implementation constraints. Inspect sensitivity to model error and parameter assumptions. Optimisation can exploit a small modelling weakness very effectively, so a surprisingly good result deserves physical scrutiny.
Translate the chosen candidate into operable settings or a buildable staged design. Record the anticipated benefit, assumptions, fallback and monitoring needed to confirm the outcome. Connect this workflow to network option assessment, staged investment and hydraulic checks.
Sources & further reading
- EPANET 2.2 User Manual ↗US Environmental Protection Agency · 2020
External source · Checked 24 September 2026 - Best Practice Modelling Guidelines ↗eWater · 2011
External source · Checked 24 September 2026
Source findings are distinguished from editorial interpretation. Apply current local criteria and project evidence when making engineering decisions.