This paper describes the application of stochastic search techniques to the production scheduling of a group of linked oil and gas fields. The goal was the maximisation of total net present value and a genetic algorithm using problem-specific crossover operators was particularly successful in this respect.


Reservoir engineers must decide upon a "best" management strategy for the exploitation of each petroleum resource, typically the strategy which will maximise economic return. In particular, a production schedule, specifying rates of extraction over the lifetime of the reservoir(s), must be chosen. From this, decisions about the construction of processing facilities and pipelines will follow. A quantitative approach to the search for the "best" production schedule requires the construction of a mathematical model, which must capture enough characteristics of the reservoir(s) to predict the costs and benefits of any given schedule.

Even when a single independent field is under consideration, finding the "best" schedule can be a non-trivial problem, as all but the simplest models are likely to be non-linear and contain discontinuities. BP has a number of multi-field potential developments world-wide. In such cases, the problem is more difficult still. The total number of production rates is of course proportionately larger, but extra complexity arises from the inter-dependence of the fields. There are extra degrees of freedom, namely the respective timing of each field, and in addition the costs associated with each field depend on the sum of the production rates from some or all of the other fields.

If possible, the reservoir engineer will also wish to ensure robustness of the choice of extraction strategy in the presence of uncertainty in the economic model - the most obvious source of uncertainty in this case being the reservoir characterisations.

This paper describes how several techniques - genetic algorithms in particular - were used to search the space of possible production schedules based on an indicative data set from one multi- field development. The techniques show potential ways to achieve significant increases in net present value, which were consistent with traditional discrete analysis.

This content is only available via PDF.
You can access this article if you purchase or spend a download.