Objectives/Scope: This paper describes a new network-solver algorithm, "GALE" ("Globally Approximate, Locally Exact") that can be used in integrated asset models that represent groups of reservoirs, wells and production facilities. The method optimizes production subject to rate, pressure, velocity, gas-lift and composition constraints and has been used to forecast and optimize very large production networks (more than 1000 wells).

The foundation of the method is iteration between solving a linear approximation of the whole production network and solving exactly the non-linear pressure calculations for individual wells and nodes. The linear global problem was solved using an object-oriented implementation of the classic simplex method, using Neumaier arithmetic to minimize round-off error. The local non-linear problems were solved with the secant method.

Tests of the GALE algorithm on both large and small models suggest that the GALE algorithm runs much faster than existing algorithms and gives results of equal or better reliability

  • A test case three-well, two-node, single reservoir, model, with monthly time-steps over 10 years, took 1 sec to run (200x faster than a "rule-based network solver" heuristic and 900x faster than an older, sequential quadratic programming (SQP) algorithm, which encountered solution problems).

  • A real-world 361-well, 77-platform model took 0.7 secs per time-step (approximately 1000x faster than an older SQP algorithm). When a larger version of this model was run with approximately 900 wells, it took 1.4 secs per time-step.

  • The GALE algorithm also appears to run somewhat (perhaps 5x) faster than the more recent non-linear optimization algorithm implemented by Wadsley.

While it is not yet known how much of the improvements are due specifically to the GALE algorithm and how much are due to other differences in the software systems, the improved calculation speeds allow IAM calculations to be applied where older algorithms would give impractically slow results (e.g. short-term operational forecasting for very big networks).

You can access this article if you purchase or spend a download.