Capacity Increase Screening (wells, subsea, topside)

Use this skill when the question is "which ideas lift or accelerate production on this host, and which should be matured?" It turns a calibrated host model into a ranked idea register. It does not replace the equipment-specific skills it calls.

Workflow

1. Baseline. Build or reuse the host model on a plan profile (oil, gas, water per year). Close mass balance and compare with plant data before any lever is tested (neqsim-process-modeling, neqsim-model-calibration-and-data-reconciliation).
2. Constraint map. Register the limit of every unit on the oil, gas, water and power paths and label its basis (vendor design case, code, operator monitoring limit, demonstrated peak). Run the year grid and record utilisation per constraint, not only the maximum (neqsim-capacity-and-utilization-analysis; ProcessModelDebottleneckStudy, BottleneckTracker).
3. Binding sequence. List which constraint binds in which years and what the next one is. A lever that relieves constraint A only buys time until constraint B binds.
4. Idea long list. Fill the lever catalogue below with the asset's own ideas first, then add missing families. Ask the asset for design margins, planned shutdowns and earlier studies before ranking.
5. Lever runs. Express each idea as a change to the model or its feed, rerun the year grid, and record the new binding sequence and the volume or time gained.
6. Register and score. One row per idea with scope, cost class, emissions, HSE, value and maturity (schema below). Test combinations, since levers rarely add.
7. Maturation pipeline and host ullage. Assign each idea a next stage and report the per-year ullage left on every constraint for future tie-ins (TieInCapacityPlanner, HostFacility).

Lever catalogue

Name a family as screened only when the model rests on assumed constants rather than tests. The completion and separation classes below are screening models: the device constants and the carry-over fractions must come from vendor flow-loop or qualification data before the result is used for more than ranking.

NeqSim classes for the well and subsea levers

• InflowControlDevice (neqsim.process.equipment.reservoir): pressure drop of an ICD (nozzle), AICD (dP = a rho_mix^2/rho_cal (mu_cal/mu_mix)^y q^x, calibrated on a reference point), AICV (closes on low viscosity) and DAR (density-window restrictor; the window and residual opening are inputs because the characteristic is vendor specific), and the inverse flow for a given pressure drop.
• InflowControlCompletion: zones with productivity index and in-situ water and gas fractions sharing one drawdown. compareAtSameOil(bareDrawdown, maxDrawdown) returns the extra drawdown the device needs for the same oil and the water and gas reductions at that oil; isOilRateReached() is false when the limit stops it. The phase split of a zone is fixed, so a coning-driven benefit is an upper bound. Use the water and gas reductions as the low/base/high lever in the host model.
• SubseaSeparationStation (neqsim.process.equipment.subsea): ThreePhaseSeparator with liquid and water pumps; set setWaterRemovalEfficiency, setOilInWaterFraction, setGasCarryUnderFraction, setLiquidExportPressure, setWaterInjectionPressure. Read getGasOutStream(), getLiquidOutStream() (to the host), getWaterOutStream(), getWaterRemovedFraction() and getTotalPowerKW(). Feed the liquid and gas streams to the host model in place of the well stream, and put the power in the emissions and electrical-load check.

Other stations (compression, hydrate and slug control, power from shore) are outside these classes; SubseaBooster covers boosting.

Rules that keep the ranking honest

• Model the lever, not the outcome. An AICD or DAR lever is an assumed shift in water and gas per barrel of oil. Report low, base and high shifts and the break-even shift at which the idea stops paying; never a single point.
• Binding metric switches. A vessel can switch binding metric (water, then oil, then gas) when the feed changes. Fit responses per (unit, metric) and rerun after every lever.
• Calibrate on the plant before using a window. If an operating window includes anti-surge recycle, a calibration of K or power against total flow overstates the future. Calibrate on net flow and keep the recycle share separate.
• Show two limit bases. Vendor or code limit and operator monitoring limit can differ in both directions. Present utilisation on both and say which one the decision uses.
• Annual mean hides peaks. Run mean and p90-day rates; exceedances occur on peak days.
• Surge and minimum-flow rows are turndown indicators, not capacity rows.
• Value needs a timeline. Acceleration value is earlier barrels, not only more barrels. Use the plan profile with and without the lever and discount both (neqsim-field-economics).
• Combinations are not additive. Run the pair and the triple of the top ideas. Report the interaction. With up to about six levers, evaluate every combination on the utilisation algebra (not new model runs) and give each lever its Shapley share of the combined gain.
• The plan profile is already capacity-capped. Count a lever's gain only in years where the base utilisation of the relieved constraint is near 1 (for example 0.85 or more), and state where the unconstrained potential comes from. If it is missing, that is the main data gap, since it sets the value scale.
• Scale rate-dependent case inputs. Wash water, lift gas and similar fixed inputs in a case runner give mass imbalance at low rates; scale them with rate and check the mass balance in the first and last year.
• Record what is ruled out and why (for example a lever that moves a constraint that is not binding), so it is not re-proposed.
• Do not let the model rank ideas it cannot see. If the dominant limit is outside the model (produced water, flare, power), screen it against demonstrated peaks and label it so.
• Match the subsea option to the binding stream. Water separation helps only when water or liquid binds; when a gas-side cluster binds (gas scrubber K, contactor momentum, injection power), screen gas-liquid separation with local gas reinjection as the subsea alternative, with the compression power from a Compressor on the separated gas, and compare it with the topside package it would replace.
• Re-run the saved valuation before reusing a task. Rebuild results.json from the scripts; a copied task can hold results from an older script version (one such rerun moved an all-levers NPV by 20 %). Check that valuation scripts still run after any edit, since a line merge can hide an elif inside a comment.
• Nameplate times regularity can be below the booked plan. For a water-flooded host the plan load for liquid, water and gas incl. lift gas often exceeds nameplate times production efficiency (one NCS FPSO: booked liquid 42.5 kSm3/d against 40 x 0.91). Test each constraint (oil, liquid, water, gas_incl_lift) against the plan peak and use the larger of the two as plan-consistent capacity; otherwise ullage is zero for decades and every tie-in looks impossible. State the capacity basis as an asset-team input.
• Ullage is per constraint and the binding one moves. Compute free ullage for oil, liquid, water and gas incl. lift separately; the binding constraint typically moves oil, then liquid, then gas incl. lift, then water as the water cut grows. A new tie-in barrel is limited by the tightest constraint at its own water-oil and gas-oil ratio, not by oil ullage.
• Cheapest access first: spare slots and host-platform wells before a new template. For an infill or second-target request, always value three alternatives on the same profile model before the hub: wells from spare slots on existing templates, wells from the host platform (free slots, sidetracks), and the new template. In the Breidablikk 5th template study the hub was at NPV P50 about zero with P(NPV<0) about 50 %, spare slots +374 and platform wells +388 MNOK (P(NPV<0) 7 % and 2 %): the hub only has a case for the targets the cheaper routes cannot reach. Slot status and reach are the first data request.
• Gas intensity per barrel is the currency of a shared LP system. LP gas per Sm3 oil = GOR + lift GLR; a heavy-oil, high-water target (Grane Nord: GOR 76 + GLR 152) needs 2.2 times the LP gas of a light depletion target (Breidablikk: 27 + 76). Compute it per well group before ranking targets, and let the stream mix set the curtailment.
• Late tie-ins displace booked barrels. Net new oil is the new production minus the base barrels displaced in the binding constraint; report the displaced share by first-oil year (one case: 43 % for first oil in 2030, 5 % or less after 2036) and the first-oil year that maximises P50 value, which is later than the earliest technical date.

Idea register schema

One row per idea; keep it in the task folder and in results.json (key_results.idea_register).

Score on the five criteria (scope, cost, emissions, HSE, value) with stated weights, and show the ranking under at least two weight sets so a reader can see whether it is stable.

Maturation pipeline

Typical stage-2 questions for the asset: measured design margins, vendor datasheet case, planned turnaround dates, installed spare slots, power headroom, past debottlenecking scope, and well-test evidence for completion ideas.

Host readiness for future tie-ins

After the lever runs, report for each year and constraint: remaining ullage on oil, gas, water, power and the unit that binds. Hand the result to TieInCapacityPlanner or the area-development workflow with the capacity basis and plan version. State which ideas must be installed before a given tie-in and which only add margin.

Deliverables

• Constraint map and binding sequence on the year grid, with the limit basis per constraint.
• Lever results per idea and for the top combinations.
• The idea register, the ranking under two weight sets and the pipeline view.
• Host ullage table.
• Assumptions, data gaps, and a list of asset questions. Fill results.json assumptions, data_gaps and agent_workflow_plan.

Related skills

• neqsim-capacity-and-utilization-analysis - constraints, snapshots, tie-in bottleneck study.
• neqsim-integrated-production-and-lifecycle - well deliverability, gas-lift allocation, tie-in planner, lifecycle.
• neqsim-subsea-and-wells - subsea equipment, well design, SURF cost.
• neqsim-field-development and neqsim-field-economics - concept comparison and NPV.
• neqsim-equipment-cost-estimation - AACE-class cost.
• neqsim-optimization-and-doe and neqsim-agentic-process-optimization - search and closed-loop setpoints.
• neqsim-professional-reporting - how the result is written up.

Gated concepts and allowable CAPEX (design-to-cost)

For re-use concepts (slot recovery, satellites on existing templates, re-use of hardware) the value depends on availability gates: line or umbilical capacity, slot fitness, spare hardware. Model each gate as a Bernoulli draw with a named fallback concept (and a delay), report the NPV distribution unconditionally and conditional on the gates passing, and include the gate as a pass/fail swing in the tornado. For design-to-cost, bisect a CAPEX scale factor on fixed Monte Carlo draws until the P50 NPV is zero (and the 30th percentile zero for a 70 % chance of positive value), and report the allowable CAPEX, the required cut and the break-even EUR per well. Worked example: the HEP reset task (task_solve, 2026-10-09 HEP VPbo): template concepts need 29-37 % lower CAPEX while slot-recovery sidetracks are positive when the line and slot gates pass.