NCS Value-Chain Optimisation

This skill optimises the Norwegian Continental Shelf as one system. It takes the open-data export graph from neqsim-ncs-infrastructure-network, puts supply on it for every year, and answers:

• how much each field and discovery should produce each year to maximise value when all fields share the same trunklines, plants and terminals;
• which elements bind, and what one more MSm3/d or Sm3/d of capacity is worth there (shadow price);
• how an outage, capacity change, price shift or new tie-in propagates;
• where spare capacity (ullage) opens up over time, and which discoveries use it best;
• which producing fields could ship more, and whether the field or the network limits them.

Everything is public screening data. Governed inputs (RNB forecasts, Gassled tariffs, gas-quality specs, metered flows) enter through a scenario JSON produced by the enterprise binding skill enterprise-ncs-value-chain-data-binding. This skill never reads company systems.

When to Use

• "Maximise NCS production/value for 2026–2040 given the pipeline system."
• "What is the most valuable debottleneck on the NCS, and when does it bind?"
• "If Kollsnes runs at 70% in 2027, what is lost and where does gas reroute?"
• "Which discoveries should be tied in first, into which host, and is there ullage?"
• "Which fields could produce more today, and what limits them?"
• Before a detailed NeqSim host process-model study: to get the rate targets the host must handle, including tie-ins.

Inputs

• An NcsNetwork (bundled snapshot or a live rebuild).
• A scenario (ncs_value_chain_scenario.v1, see scenario.py) with these keys:
  • years;
  • prices (gas NOK/Sm3, liquid NOK/Sm3, exit-market factors);
  • tariffs (default and per arc);
  • discount_rate;
  • capacity_overrides;
  • outages (element, years, available fraction);
  • forecast_overrides (e.g. RNB per field);
  • include_discoveries, tieback_max_km, max_paths_per_entity. All have stated public defaults.

Outputs

NcsValueChainOptimizer(network, scenario).run() returns ncs_value_chain_result.v1. For each year it holds:

• produced per entity (u, gas MSm3/d, liquid Sm3/d);
• curtailed, element_flow and utilization;
• shadow_price_mnok_per_unit_yr, and exits by market;
• unsold (production without a sales route, e.g. reinjection) and unconnected entities.

It also carries the horizon NPV, the tie-back host of each discovery, the supply profiles with their method, and the stated assumptions.

opportunities.py turns this into:

• bottlenecks: elements ranked by discounted shadow value;
• ullage_timeline: spare capacity per element and year;
• curtailment: lost production by year and entity;
• tiein_ranking: discoveries by accepted discounted revenue, with host and distance;
• stranded_discoveries: no host within the tie-back radius;
• production_uplift: extra rate per field, limited by route headroom or demonstrated capability.

Engineering Method

1. Supply. Sodir yearly net production up to the base year.
   • Forecast: an Arps decline fitted from peak, using norwegian_continental_shelf_data.fit_arps_decline (built-in exponential fallback), with a 4 %/yr minimum decline. Fields still on plateau hold their rate until about 60 % of remaining reserves are produced, then decline at 12 %/yr.
   • Cap: cumulative future volume is capped at Sodir remaining reserves.
   • Discoveries: sized from Sodir recoverable volumes. Start year = base year + 2 (decided/approved), + 4 (clarification), + 7 (likely); not evaluated and unlikely are excluded by default. Profile: 50 % build-up year, then 12 %/yr of recoverable on plateau for 4 years, then 15 %/yr decline.
2. Routing. Up to 20 loop-free paths per entity and medium to markets. A discovery ties back to the nearest in-service processing host within tieback_max_km that has an export route.

3. LP per year.

   $$\max \sum_{e,p} \left(P_m f_{exit} - T_p\right) x_{e,p} \quad \text{s.t.} \quad \sum_p x^{m}{e,p} = u_e R^{m}e,; \sum{(e,p) \ni c} x{e,p} \le C_c,; 0 \le u_e \le 1$$

   Here $u_e$ couples a field's oil and gas, and $C_c$ is every trunkline and plant capacity after overrides and outages. Solved with SciPy HiGHS. Duals of the capacity rows are shadow prices in MNOK/yr per MSm3/d (gas) or per Sm3/d (liquid).

4. Horizon value. Annual objectives discounted at discount_rate. Years are independent: curtailed volume is lost, not deferred (conservative).

Python Usage Pattern

from ncs_infrastructure_network import NcsNetwork
from ncs_value_chain_optimization import (NcsValueChainOptimizer, make_scenario, bottlenecks,
                                          tiein_ranking, production_uplift, ullage_timeline)
from ncs_value_chain_optimization import neqsim_bridge as vb

net = NcsNetwork.load()
sc = make_scenario(years=[2026, 2040],
                   outages=[{"element": "KOLLSNES", "years": [2027], "available_fraction": 0.7}])
opt = NcsValueChainOptimizer(net, sc)
res = opt.run()                           # ~1 s for 15 years, 160 entities

bottlenecks(res)[:5]                      # KOLLSNES binds in 2027 at ~1090 MNOK/yr per MSm3/d
tiein_ranking(res, gas_price=3.0, liquid_price=4500.0)[:10]
production_uplift(opt, res, 2026)
ullage_timeline(res, ["LANGELED_SOUTH__SLEIPNER__EASINGTON", "KOLLSNES"])

# NeqSim hand-offs (pip install neqsim)
vb.value_objective_check(res, sc, 2026)   # ValueChainObjective reproduces the LP gross revenue
specs = vb.debottleneck_specs(opt, res, [{"element": "KOLLSNES", "added_capacity": 10, "capex_mnok": 3000}])
vb.run_debottlenecking(specs, sc)         # DebottleneckingAdvisor: NPV, B/C, payback
vb.process_model_targets(res, net, "GOLIAT")  # rate targets for the optimize-processmodel agent

Validated NeqSim Path

Validation Checklist

• Every year status == "optimal"; first-year shipped gas within a few % of the last historical year (the 2026 base case ships ~320 MSm3/d).
• max(utilization) <= 1 in every year.
• value_objective_check relative difference below 1e-6 (the NeqSim economics and the LP agree).
• Shadow prices ≈ netback × 365 on binding gas elements (sanity: (3.0 − 0.12) × 365 ≈ 1051 MNOK/yr per MSm3/d).
• Discoveries used in rankings carry status and RC; start years reviewed.
• Prices and tariffs are labelled public defaults unless bound from governed data.

Common Mistakes

• Too few paths per entity. With 8 paths the LP curtailed about 19 MSm3/d of Troll gas that the network can carry. Keep ≥ 20.
• Treating DebottleneckCandidate years as calendar years. They are offsets from the base year; 2027 as a year discounts the benefit to zero.
• Using duals for large capacity steps. A dual is marginal. The bridge re-solves with the added capacity instead.
• Reading profiles as a forecast. Sodir layer 7300 is history only; the current year is year-to-date.
• Mistaking zero curtailment for no bottleneck. When supply declines faster than capacity, nothing binds. Check ullage_timeline for where tie-ins fit.

Limitations

• Nameplate capacities; no line-pack, pressure-dependent capacity, gas-quality blending, Gassco nomination rules or NGL product constraints at Kårstø.
• Screening supply forecasts (decline + reserves cap). Replace them with RNB or operator forecasts through forecast_overrides for decisions.
• Tariffs and exit-market factors are placeholders until bound from governed commercial data.
• Years are independent (no deferral, storage or reservoir response to curtailment).
• Host processing capacity is not open data; the host process model must confirm tie-in rates.

Related NeqSim Functionality

• neqsim-ncs-infrastructure-network: the graph, open-API clients, tie-back and hydraulics bridges.
• neqsim-norwegian-continental-shelf-data: fit_arps_decline, headline facts, carbon-cost basis.
• neqsim-resource-classification-screening: RC maturity behind discovery phasing.
• neqsim-asset-value-npv-screening, neqsim-energy-emissions-screening: project-level economics and emissions of a chosen tie-in or debottleneck.
• NeqSim optimize-processmodel agent: host plant confirmation of the rate targets.

References

• Norwegian Offshore Directorate FactPages / DataService (NLOD 2.0).
• Norwegian Petroleum, "The oil and gas pipeline system" and "Resource accounts".
• Arps, J.J. (1945). Analysis of decline curves. Trans. AIME 160, 228–247.
• Huangfu, Q., Hall, J.A.J. (2018). Parallelizing the dual revised simplex method. Math. Prog. Comp. 10, 119–142 (HiGHS).